Dataset Viewer
Auto-converted to Parquet Duplicate
id
stringlengths
12
32
description
stringlengths
16
72
prompt
stringlengths
39
249
servers
listlengths
1
1
setup
listlengths
0
5
check
dict
tags
listlengths
2
4
difficulty
stringclasses
3 values
skill
stringclasses
3 values
server
stringclasses
6 values
check_json
stringlengths
62
348
setup_json
stringlengths
2
356
fetch_001_title
Fetch a page and identify its title text
Fetch the page at https://example.com and tell me what heading or title appears on it.
[ "fetch" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "Example Domain" } }
[ "single-tool", "read-only" ]
easy
selection
fetch
{"kind": "final_text_contains", "params": {"substring": "Example Domain"}}
[]
fetch_002_json_field
Fetch a JSON endpoint and extract a field
Fetch the JSON at https://httpbin.org/json and tell me the value of the 'title' field of the slideshow.
[ "fetch" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "Sample Slide Show" } }
[ "single-tool", "read-only", "json" ]
easy
selection
fetch
{"kind": "final_text_contains", "params": {"substring": "Sample Slide Show"}}
[]
fetch_003_two_pages
Fetch two pages and answer a question about both
Fetch https://example.com and https://www.iana.org/help/example-domains. Do both pages mention the word 'example'? Answer yes or no.
[ "fetch" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "yes" } }
[ "multi-step", "read-only" ]
medium
composition
fetch
{"kind": "final_text_contains", "params": {"substring": "yes"}}
[]
fetch_004_recover_invalid_host
Error recovery after a host that does not resolve
Try to fetch http://this-domain-does-not-exist-987654321.invalid/ first. If it fails, fetch https://example.com instead and tell me what heading is on that page.
[ "fetch" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "Example Domain" } }
[ "error-recovery", "multi-step", "RQ2" ]
medium
recovery
fetch
{"kind": "final_text_contains", "params": {"substring": "Example Domain"}}
[]
fetch_005_uuid_format
Fetch a UUID endpoint and report the value
Fetch https://httpbin.org/uuid and tell me the UUID value that the response contains.
[ "fetch" ]
[]
{ "kind": "final_text_regex", "params": { "pattern": "[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}" } }
[ "single-tool", "read-only", "json" ]
easy
selection
fetch
{"kind": "final_text_regex", "params": {"pattern": "[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}"}}
[]
fetch_006_keyword_presence
Fetch a page and check for a specific word
Fetch https://example.com. Does the page body contain the word 'illustrative'? Answer yes or no.
[ "fetch" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "yes" } }
[ "single-tool", "read-only" ]
easy
selection
fetch
{"kind": "final_text_contains", "params": {"substring": "yes"}}
[]
fetch_007_link_destination
Fetch a page and identify the link it points to
Fetch https://example.com. The page contains exactly one hyperlink in its body. Tell me what domain that link points to.
[ "fetch" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "iana.org" } }
[ "single-tool", "read-only" ]
medium
selection
fetch
{"kind": "final_text_contains", "params": {"substring": "iana.org"}}
[]
fetch_008_json_nested
Fetch a JSON document and navigate a nested field
Fetch https://httpbin.org/json. In the slideshow object, the slides array has multiple entries. Tell me the 'type' of the first slide.
[ "fetch" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "all" } }
[ "single-tool", "read-only", "json" ]
medium
selection
fetch
{"kind": "final_text_contains", "params": {"substring": "all"}}
[]
fetch_009_recover_http_500
Error recovery after a non-2xx HTTP response
Fetch https://httpbin.org/status/500. If the request fails or returns an error, fetch https://example.com instead and tell me the page's main heading.
[ "fetch" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "Example Domain" } }
[ "error-recovery", "multi-step", "RQ2" ]
medium
recovery
fetch
{"kind": "final_text_contains", "params": {"substring": "Example Domain"}}
[]
fetch_010_dual_source
Fetch two endpoints and combine the result
Fetch both https://httpbin.org/uuid and https://httpbin.org/json. Tell me both: the uuid value, and the title of the slideshow.
[ "fetch" ]
[]
{ "kind": "all_of", "params": { "checks": [ { "kind": "final_text_regex", "params": { "pattern": "[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}" } }, { "kind": "final_text_contains", "params": { "substring": "Sample Sli...
[ "multi-step", "read-only", "json" ]
hard
composition
fetch
{"kind": "all_of", "params": {"checks": [{"kind": "final_text_regex", "params": {"pattern": "[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}"}}, {"kind": "final_text_contains", "params": {"substring": "Sample Slide Show"}}]}}
[]
fetch_011_host_header
Fetch a headers endpoint and extract one header
Fetch https://httpbin.org/headers. The response is JSON listing the request headers. Tell me the value of the 'Host' header.
[ "fetch" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "httpbin.org" } }
[ "single-tool", "read-only", "json" ]
medium
selection
fetch
{"kind": "final_text_contains", "params": {"substring": "httpbin.org"}}
[]
fetch_012_count_slides
Fetch JSON and count array elements
Fetch https://httpbin.org/json. How many slides are in the slideshow's 'slides' array? Tell me the number.
[ "fetch" ]
[]
{ "kind": "final_text_regex", "params": { "pattern": "\\b2\\b" } }
[ "single-tool", "read-only", "json" ]
medium
selection
fetch
{"kind": "final_text_regex", "params": {"pattern": "\\b2\\b"}}
[]
fetch_013_combine_sources
Fetch a page and a JSON endpoint, report both
Fetch https://example.com and https://httpbin.org/uuid. Tell me both the heading on example.com and the uuid value from the other endpoint.
[ "fetch" ]
[]
{ "kind": "all_of", "params": { "checks": [ { "kind": "final_text_contains", "params": { "substring": "Example Domain" } }, { "kind": "final_text_regex", "params": { "pattern": "[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]...
[ "multi-step", "read-only" ]
hard
composition
fetch
{"kind": "all_of", "params": {"checks": [{"kind": "final_text_contains", "params": {"substring": "Example Domain"}}, {"kind": "final_text_regex", "params": {"pattern": "[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}"}}]}}
[]
fs_001_read_line
Read a file and report a specific line
Read the file 'notes.txt' in your working directory and tell me what the third line says.
[ "filesystem" ]
[ { "kind": "write_file", "params": { "path": "notes.txt", "content": "line one is here\nline two is here\nthe third line says: orange marmalade\nline four\n" } } ]
{ "kind": "final_text_contains", "params": { "substring": "orange marmalade" } }
[ "single-tool", "read-only" ]
easy
selection
filesystem
{"kind": "final_text_contains", "params": {"substring": "orange marmalade"}}
[{"kind": "write_file", "params": {"path": "notes.txt", "content": "line one is here\nline two is here\nthe third line says: orange marmalade\nline four\n"}}]
fs_002_write_exact
Write a file with exact content
Create a file named 'greeting.txt' in your working directory containing exactly the text: Hello, MCP!
[ "filesystem" ]
[]
{ "kind": "file_content_equals", "params": { "path": "greeting.txt", "content": "Hello, MCP!", "strip": true } }
[ "single-tool", "write" ]
easy
selection
filesystem
{"kind": "file_content_equals", "params": {"path": "greeting.txt", "content": "Hello, MCP!", "strip": true}}
[]
fs_003_count_txt
List a directory and count files by extension
Look at the files in your working directory. How many of them have the .txt extension?
[ "filesystem" ]
[ { "kind": "write_file", "params": { "path": "a.txt", "content": "a" } }, { "kind": "write_file", "params": { "path": "b.txt", "content": "b" } }, { "kind": "write_file", "params": { "path": "c.txt", "content": "c" } }, { "kind":...
{ "kind": "final_text_regex", "params": { "pattern": "\\b3\\b" } }
[ "single-tool", "read-only", "counting" ]
easy
selection
filesystem
{"kind": "final_text_regex", "params": {"pattern": "\\b3\\b"}}
[{"kind": "write_file", "params": {"path": "a.txt", "content": "a"}}, {"kind": "write_file", "params": {"path": "b.txt", "content": "b"}}, {"kind": "write_file", "params": {"path": "c.txt", "content": "c"}}, {"kind": "write_file", "params": {"path": "data.csv", "content": "x,y"}}, {"kind": "write_file", "params": {"pat...
fs_004_compare_files
Read two files and compare their content
Read the files 'a.txt' and 'b.txt'. Do they contain the same content? Answer yes or no.
[ "filesystem" ]
[ { "kind": "write_file", "params": { "path": "a.txt", "content": "shared content here" } }, { "kind": "write_file", "params": { "path": "b.txt", "content": "shared content here" } } ]
{ "kind": "final_text_contains", "params": { "substring": "yes" } }
[ "multi-step", "read-only" ]
medium
composition
filesystem
{"kind": "final_text_contains", "params": {"substring": "yes"}}
[{"kind": "write_file", "params": {"path": "a.txt", "content": "shared content here"}}, {"kind": "write_file", "params": {"path": "b.txt", "content": "shared content here"}}]
fs_005_find_file
Search for a file by name pattern in a directory tree
There is a file somewhere in this directory tree whose name contains the word 'secret'. Find it and tell me its name (just the filename, not the full path).
[ "filesystem" ]
[ { "kind": "mkdir", "params": { "path": "docs/2024" } }, { "kind": "mkdir", "params": { "path": "logs" } }, { "kind": "write_file", "params": { "path": "docs/readme.txt", "content": "doc" } }, { "kind": "write_file", "params": { "p...
{ "kind": "final_text_contains", "params": { "substring": "secret_notes" } }
[ "multi-step", "read-only", "search" ]
medium
composition
filesystem
{"kind": "final_text_contains", "params": {"substring": "secret_notes"}}
[{"kind": "mkdir", "params": {"path": "docs/2024"}}, {"kind": "mkdir", "params": {"path": "logs"}}, {"kind": "write_file", "params": {"path": "docs/readme.txt", "content": "doc"}}, {"kind": "write_file", "params": {"path": "logs/run.log", "content": "log"}}, {"kind": "write_file", "params": {"path": "docs/2024/secret_n...
fs_006_edit_in_place
Edit a file by replacing one word
There is a file 'report.md' in your working directory. Replace the word 'TODO' with the word 'DONE' in that file. Keep the rest of the content unchanged.
[ "filesystem" ]
[ { "kind": "write_file", "params": { "path": "report.md", "content": "# Report\n\nStatus: TODO\n\nDetails follow.\n" } } ]
{ "kind": "file_content_equals", "params": { "path": "report.md", "content": "# Report\n\nStatus: DONE\n\nDetails follow.\n" } }
[ "single-tool", "write", "edit" ]
medium
selection
filesystem
{"kind": "file_content_equals", "params": {"path": "report.md", "content": "# Report\n\nStatus: DONE\n\nDetails follow.\n"}}
[{"kind": "write_file", "params": {"path": "report.md", "content": "# Report\n\nStatus: TODO\n\nDetails follow.\n"}}]
fs_007_create_nested
Create a directory and place a file in it
Create a directory named 'out' in your working directory, and inside it create a file 'summary.txt' that contains the word: done
[ "filesystem" ]
[]
{ "kind": "all_of", "params": { "checks": [ { "kind": "file_exists", "params": { "path": "out/summary.txt" } }, { "kind": "file_content_contains", "params": { "path": "out/summary.txt", "substring": "done" } } ...
[ "multi-step", "write" ]
medium
composition
filesystem
{"kind": "all_of", "params": {"checks": [{"kind": "file_exists", "params": {"path": "out/summary.txt"}}, {"kind": "file_content_contains", "params": {"path": "out/summary.txt", "substring": "done"}}]}}
[]
fs_008_rename
Rename a file, preserving content
Rename the file 'old_name.txt' to 'new_name.txt'. Do not change its content.
[ "filesystem" ]
[ { "kind": "write_file", "params": { "path": "old_name.txt", "content": "important payload" } } ]
{ "kind": "all_of", "params": { "checks": [ { "kind": "file_not_exists", "params": { "path": "old_name.txt" } }, { "kind": "file_content_equals", "params": { "path": "new_name.txt", "content": "important payload" } ...
[ "single-tool", "write" ]
easy
selection
filesystem
{"kind": "all_of", "params": {"checks": [{"kind": "file_not_exists", "params": {"path": "old_name.txt"}}, {"kind": "file_content_equals", "params": {"path": "new_name.txt", "content": "important payload"}}]}}
[{"kind": "write_file", "params": {"path": "old_name.txt", "content": "important payload"}}]
fs_009_recover_missing_file
Error recovery: read a missing file, then create it
First, try to read the file 'does_not_exist.txt'. If reading it fails because the file is missing, then create that file with the content: placeholder
[ "filesystem" ]
[]
{ "kind": "file_content_equals", "params": { "path": "does_not_exist.txt", "content": "placeholder", "strip": true } }
[ "error-recovery", "multi-step", "RQ2" ]
medium
recovery
filesystem
{"kind": "file_content_equals", "params": {"path": "does_not_exist.txt", "content": "placeholder", "strip": true}}
[]
fs_010_count_lines
Read multiple files and sum a numeric property
There are several .log files in your working directory. Count the total number of lines across all of them and tell me the total. (A line is anything ending in a newline.)
[ "filesystem" ]
[ { "kind": "write_file", "params": { "path": "a.log", "content": "l1\nl2\nl3\n" } }, { "kind": "write_file", "params": { "path": "b.log", "content": "l1\nl2\n" } }, { "kind": "write_file", "params": { "path": "c.log", "content": "l1\nl2\nl3\...
{ "kind": "final_text_regex", "params": { "pattern": "\\b10\\b" } }
[ "multi-step", "read-only", "counting" ]
hard
composition
filesystem
{"kind": "final_text_regex", "params": {"pattern": "\\b10\\b"}}
[{"kind": "write_file", "params": {"path": "a.log", "content": "l1\nl2\nl3\n"}}, {"kind": "write_file", "params": {"path": "b.log", "content": "l1\nl2\n"}}, {"kind": "write_file", "params": {"path": "c.log", "content": "l1\nl2\nl3\nl4\nl5\n"}}]
fs_011_concat_files
Read multiple files and write their concatenation
There are three text files (a.txt, b.txt, c.txt) in your working directory. Read all of them and write a new file 'combined.txt' that contains the contents of all three.
[ "filesystem" ]
[ { "kind": "write_file", "params": { "path": "a.txt", "content": "apple\n" } }, { "kind": "write_file", "params": { "path": "b.txt", "content": "banana\n" } }, { "kind": "write_file", "params": { "path": "c.txt", "content": "cherry\n" } ...
{ "kind": "all_of", "params": { "checks": [ { "kind": "file_content_contains", "params": { "path": "combined.txt", "substring": "apple" } }, { "kind": "file_content_contains", "params": { "path": "combined.txt", "s...
[ "multi-step", "write" ]
hard
composition
filesystem
{"kind": "all_of", "params": {"checks": [{"kind": "file_content_contains", "params": {"path": "combined.txt", "substring": "apple"}}, {"kind": "file_content_contains", "params": {"path": "combined.txt", "substring": "banana"}}, {"kind": "file_content_contains", "params": {"path": "combined.txt", "substring": "cherry"}}...
[{"kind": "write_file", "params": {"path": "a.txt", "content": "apple\n"}}, {"kind": "write_file", "params": {"path": "b.txt", "content": "banana\n"}}, {"kind": "write_file", "params": {"path": "c.txt", "content": "cherry\n"}}]
fs_012_largest_file
Identify the largest file by size
Several files are in your working directory. Tell me the name of the file that has the largest size in bytes.
[ "filesystem" ]
[ { "kind": "write_file", "params": { "path": "tiny.txt", "content": "x" } }, { "kind": "write_file", "params": { "path": "biggest.txt", "content": "this file has by far the most content of all the files in this directory, by a wide margin indeed" } }, { "ki...
{ "kind": "final_text_contains", "params": { "substring": "biggest.txt" } }
[ "multi-step", "read-only" ]
medium
composition
filesystem
{"kind": "final_text_contains", "params": {"substring": "biggest.txt"}}
[{"kind": "write_file", "params": {"path": "tiny.txt", "content": "x"}}, {"kind": "write_file", "params": {"path": "biggest.txt", "content": "this file has by far the most content of all the files in this directory, by a wide margin indeed"}}, {"kind": "write_file", "params": {"path": "medium.txt", "content": "a modera...
fs_013_increment_counter
Read a number, compute, and write it back
The file 'counter.txt' contains a single integer. Read it, add 1 to it, and write the new value back to 'counter.txt'.
[ "filesystem" ]
[ { "kind": "write_file", "params": { "path": "counter.txt", "content": "41" } } ]
{ "kind": "file_content_equals", "params": { "path": "counter.txt", "content": "42", "strip": true } }
[ "multi-step", "write" ]
medium
composition
filesystem
{"kind": "file_content_equals", "params": {"path": "counter.txt", "content": "42", "strip": true}}
[{"kind": "write_file", "params": {"path": "counter.txt", "content": "41"}}]
fs_014_delete_by_content
Find and delete the file with specific content
Three files are in your working directory. Exactly one of them contains the text 'DELETE ME'. Find that file and delete it. Leave the others untouched.
[ "filesystem" ]
[ { "kind": "write_file", "params": { "path": "keep1.txt", "content": "important data" } }, { "kind": "write_file", "params": { "path": "target.txt", "content": "DELETE ME" } }, { "kind": "write_file", "params": { "path": "keep2.txt", "conten...
{ "kind": "all_of", "params": { "checks": [ { "kind": "file_not_exists", "params": { "path": "target.txt" } }, { "kind": "file_exists", "params": { "path": "keep1.txt" } }, { "kind": "file_exists", ...
[ "multi-step", "write" ]
hard
composition
filesystem
{"kind": "all_of", "params": {"checks": [{"kind": "file_not_exists", "params": {"path": "target.txt"}}, {"kind": "file_exists", "params": {"path": "keep1.txt"}}, {"kind": "file_exists", "params": {"path": "keep2.txt"}}]}}
[{"kind": "write_file", "params": {"path": "keep1.txt", "content": "important data"}}, {"kind": "write_file", "params": {"path": "target.txt", "content": "DELETE ME"}}, {"kind": "write_file", "params": {"path": "keep2.txt", "content": "also keep this"}}]
gh_001_default_branch
Look up the default branch of a public GitHub repository
What is the name of the default branch of the GitHub repository 'microsoft/vscode'?
[ "github" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "main" } }
[ "single-tool", "read-only" ]
easy
selection
github
{"kind": "final_text_contains", "params": {"substring": "main"}}
[]
gh_002_primary_language
Report the primary programming language of a public repo
What is the primary programming language reported by GitHub for the 'microsoft/TypeScript' repository?
[ "github" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "TypeScript" } }
[ "single-tool", "read-only" ]
easy
selection
github
{"kind": "final_text_contains", "params": {"substring": "TypeScript"}}
[]
gh_003_read_license
Fetch the LICENSE file of a repo and identify the license
Fetch the LICENSE file from the 'modelcontextprotocol/servers' repository on GitHub and tell me which type of open-source license it grants.
[ "github" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "MIT" } }
[ "single-tool", "read-only" ]
medium
selection
github
{"kind": "final_text_contains", "params": {"substring": "MIT"}}
[]
gh_004_search_org_repos
Search an organization's repositories and check for existence
Search GitHub for repositories owned by the 'anthropics' organization. Does a repository named 'prompt-eng-interactive-tutorial' appear in the results? Answer yes or no.
[ "github" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "yes" } }
[ "single-tool", "read-only", "search" ]
medium
selection
github
{"kind": "final_text_contains", "params": {"substring": "yes"}}
[]
gh_005_repo_description
Read a repository description and identify a keyword
Fetch the description of the 'modelcontextprotocol/python-sdk' repository on GitHub. Does the description mention Python? Answer yes or no.
[ "github" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "yes" } }
[ "single-tool", "read-only" ]
easy
selection
github
{"kind": "final_text_contains", "params": {"substring": "yes"}}
[]
gh_006_readme_keyword
Fetch a README and confirm a keyword is present
Fetch the README of the 'modelcontextprotocol/python-sdk' repository on GitHub. Does it mention the Model Context Protocol? Answer yes or no.
[ "github" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "yes" } }
[ "single-tool", "read-only" ]
medium
selection
github
{"kind": "final_text_contains", "params": {"substring": "yes"}}
[]
gh_007_search_code_filename
Search a repository for a file matching a keyword
In the GitHub repository 'modelcontextprotocol/servers', is there any file or directory whose name contains the word 'sqlite'? Answer yes or no.
[ "github" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "yes" } }
[ "multi-step", "read-only", "search" ]
medium
composition
github
{"kind": "final_text_contains", "params": {"substring": "yes"}}
[]
gh_008_two_repos_compare
Fetch metadata from two repos and compare a field
Look up the default branch of both 'microsoft/vscode' and 'modelcontextprotocol/servers'. Are the two default branch names identical? Answer yes or no.
[ "github" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "yes" } }
[ "multi-step", "read-only" ]
hard
composition
github
{"kind": "final_text_contains", "params": {"substring": "yes"}}
[]
gh_009_recover_invalid_repo
Error recovery: query a non-existent repo, then a real one
First, try to fetch the default branch of the repository 'this-org-does-not-exist-xyz999/nonexistent-repo'. If that fails because the repository does not exist, instead fetch the default branch of 'modelcontextprotocol/servers' and tell me its name.
[ "github" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "main" } }
[ "error-recovery", "multi-step", "RQ2" ]
medium
recovery
github
{"kind": "final_text_contains", "params": {"substring": "main"}}
[]
gh_010_search_issues_keyword
Search issues for a topical keyword and check existence
Search the issues of the 'modelcontextprotocol/python-sdk' repository on GitHub for ones that mention the word 'tool'. Does at least one such issue exist? Answer yes or no.
[ "github" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "yes" } }
[ "single-tool", "read-only", "search" ]
medium
selection
github
{"kind": "final_text_contains", "params": {"substring": "yes"}}
[]
mem_001_create_entity
Create a single entity in the knowledge graph
Store a new entity in memory named 'Alice' with the entity type 'person'.
[ "memory" ]
[]
{ "kind": "memory_has_entity", "params": { "name": "Alice", "entity_type": "person" } }
[ "single-tool", "write" ]
easy
selection
memory
{"kind": "memory_has_entity", "params": {"name": "Alice", "entity_type": "person"}}
[]
mem_002_entity_with_observation
Create an entity carrying an observation
Remember that 'Paris' is a city. Record an observation on it stating that it is the capital of France.
[ "memory" ]
[]
{ "kind": "memory_entity_observation_contains", "params": { "name": "Paris", "substring": "capital of France" } }
[ "single-tool", "write" ]
easy
selection
memory
{"kind": "memory_entity_observation_contains", "params": {"name": "Paris", "substring": "capital of France"}}
[]
mem_003_create_relation
Create two entities and a relation between them
Create an entity 'Alice' (type person) and an entity 'Acme' (type company). Then record that Alice works at Acme.
[ "memory" ]
[]
{ "kind": "all_of", "params": { "checks": [ { "kind": "memory_has_entity", "params": { "name": "Alice" } }, { "kind": "memory_has_entity", "params": { "name": "Acme" } }, { "kind": "memory_has_relation", ...
[ "multi-step", "write" ]
medium
composition
memory
{"kind": "all_of", "params": {"checks": [{"kind": "memory_has_entity", "params": {"name": "Alice"}}, {"kind": "memory_has_entity", "params": {"name": "Acme"}}, {"kind": "memory_has_relation", "params": {"from": "Alice", "to": "Acme"}}]}}
[]
mem_004_add_observation_existing
Add an observation to a pre-existing entity
There is already a person named 'Bob' in memory. Add an observation to Bob recording that he likes coffee.
[ "memory" ]
[ { "kind": "init_memory", "params": { "entities": [ { "name": "Bob", "entityType": "person", "observations": [ "lives in Seattle" ] } ] } } ]
{ "kind": "memory_entity_observation_contains", "params": { "name": "Bob", "substring": "coffee" } }
[ "single-tool", "write" ]
medium
selection
memory
{"kind": "memory_entity_observation_contains", "params": {"name": "Bob", "substring": "coffee"}}
[{"kind": "init_memory", "params": {"entities": [{"name": "Bob", "entityType": "person", "observations": ["lives in Seattle"]}]}}]
mem_005_count_entities
Read the graph and count entities
How many entities are currently stored in memory? Tell me the number.
[ "memory" ]
[ { "kind": "init_memory", "params": { "entities": [ { "name": "Alice", "entityType": "person" }, { "name": "Bob", "entityType": "person" }, { "name": "Acme", "entityType": "company" } ] ...
{ "kind": "final_text_regex", "params": { "pattern": "\\b3\\b" } }
[ "single-tool", "read-only" ]
easy
selection
memory
{"kind": "final_text_regex", "params": {"pattern": "\\b3\\b"}}
[{"kind": "init_memory", "params": {"entities": [{"name": "Alice", "entityType": "person"}, {"name": "Bob", "entityType": "person"}, {"name": "Acme", "entityType": "company"}]}}]
mem_006_search_nodes
Search the graph for a matching node
Search memory for any node related to 'Python' and tell me the name of the entity you find.
[ "memory" ]
[ { "kind": "init_memory", "params": { "entities": [ { "name": "Guido", "entityType": "person", "observations": [ "created the Python language" ] }, { "name": "Acme", "entityType": "company", "obser...
{ "kind": "final_text_contains", "params": { "substring": "Guido" } }
[ "single-tool", "read-only", "search" ]
medium
selection
memory
{"kind": "final_text_contains", "params": {"substring": "Guido"}}
[{"kind": "init_memory", "params": {"entities": [{"name": "Guido", "entityType": "person", "observations": ["created the Python language"]}, {"name": "Acme", "entityType": "company", "observations": ["sells widgets"]}]}}]
mem_007_delete_entity
Delete an entity from the graph
Delete the entity named 'Bob' from memory.
[ "memory" ]
[ { "kind": "init_memory", "params": { "entities": [ { "name": "Alice", "entityType": "person" }, { "name": "Bob", "entityType": "person" } ] } } ]
{ "kind": "all_of", "params": { "checks": [ { "kind": "memory_not_has_entity", "params": { "name": "Bob" } }, { "kind": "memory_has_entity", "params": { "name": "Alice" } } ] } }
[ "single-tool", "write" ]
medium
composition
memory
{"kind": "all_of", "params": {"checks": [{"kind": "memory_not_has_entity", "params": {"name": "Bob"}}, {"kind": "memory_has_entity", "params": {"name": "Alice"}}]}}
[{"kind": "init_memory", "params": {"entities": [{"name": "Alice", "entityType": "person"}, {"name": "Bob", "entityType": "person"}]}}]
mem_008_relation_query
Query the graph across relations
Using memory, tell me the names of everyone who works at 'Acme'.
[ "memory" ]
[ { "kind": "init_memory", "params": { "entities": [ { "name": "Alice", "entityType": "person" }, { "name": "Bob", "entityType": "person" }, { "name": "Carol", "entityType": "person" }, { ...
{ "kind": "final_text_contains", "params": { "substrings": [ "Alice", "Bob" ], "require_all": true } }
[ "multi-step", "read-only" ]
hard
composition
memory
{"kind": "final_text_contains", "params": {"substrings": ["Alice", "Bob"], "require_all": true}}
[{"kind": "init_memory", "params": {"entities": [{"name": "Alice", "entityType": "person"}, {"name": "Bob", "entityType": "person"}, {"name": "Carol", "entityType": "person"}, {"name": "Acme", "entityType": "company"}], "relations": [{"from": "Alice", "to": "Acme", "relationType": "works_at"}, {"from": "Bob", "to": "Ac...
mem_009_recover_missing_entity
Error recovery: add observation to a non-existent entity, then create it
Add the observation 'newly created' to an entity named 'Ghost'. If that fails because Ghost does not exist, create Ghost as a person first and then add the observation.
[ "memory" ]
[]
{ "kind": "memory_entity_observation_contains", "params": { "name": "Ghost", "substring": "newly created" } }
[ "error-recovery", "multi-step", "RQ2" ]
medium
recovery
memory
{"kind": "memory_entity_observation_contains", "params": {"name": "Ghost", "substring": "newly created"}}
[]
mem_010_build_chain
Build a multi-entity relation chain
Create three people named 'X', 'Y', and 'Z'. Record that X knows Y, and that Y knows Z.
[ "memory" ]
[]
{ "kind": "all_of", "params": { "checks": [ { "kind": "memory_has_entity", "params": { "name": "X" } }, { "kind": "memory_has_entity", "params": { "name": "Y" } }, { "kind": "memory_has_entity", "pa...
[ "multi-step", "write" ]
hard
composition
memory
{"kind": "all_of", "params": {"checks": [{"kind": "memory_has_entity", "params": {"name": "X"}}, {"kind": "memory_has_entity", "params": {"name": "Y"}}, {"kind": "memory_has_entity", "params": {"name": "Z"}}, {"kind": "memory_has_relation", "params": {"from": "X", "to": "Y"}}, {"kind": "memory_has_relation", "params": ...
[]
pg_001_count_employees
Count rows in a single table
Query the database: how many employees are there in total?
[ "postgres" ]
[]
{ "kind": "final_text_regex", "params": { "pattern": "\\b5\\b" } }
[ "single-tool", "read-only" ]
easy
selection
postgres
{"kind": "final_text_regex", "params": {"pattern": "\\b5\\b"}}
[]
pg_002_list_departments
Read every row of the departments table
Query the database for the names of every department. List all of them.
[ "postgres" ]
[]
{ "kind": "final_text_contains", "params": { "substrings": [ "Engineering", "Sales", "HR" ], "require_all": true } }
[ "single-tool", "read-only" ]
easy
selection
postgres
{"kind": "final_text_contains", "params": {"substrings": ["Engineering", "Sales", "HR"], "require_all": true}}
[]
pg_003_oldest_employee
Find the row with the earliest date in a column
Which employee has the earliest hire date according to the database? Tell me their name.
[ "postgres" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "Eve" } }
[ "single-tool", "read-only" ]
medium
selection
postgres
{"kind": "final_text_contains", "params": {"substring": "Eve"}}
[]
pg_004_average_salary
Aggregate query (average) across a table
Compute the average salary across all employees. Tell me the integer-rounded value.
[ "postgres" ]
[]
{ "kind": "final_text_regex", "params": { "pattern": "94,?000" } }
[ "single-tool", "read-only", "aggregate" ]
medium
selection
postgres
{"kind": "final_text_regex", "params": {"pattern": "94,?000"}}
[]
pg_005_highest_paid
Find the employee with the maximum value across a join
Which employee earns the highest salary? Tell me their name.
[ "postgres" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "Alice" } }
[ "multi-step", "read-only", "join" ]
medium
composition
postgres
{"kind": "final_text_contains", "params": {"substring": "Alice"}}
[]
pg_006_recent_sales_hire
Join with a WHERE filter and order by date
Who is the most recently hired employee in the 'Sales' department? Tell me their name.
[ "postgres" ]
[]
{ "kind": "final_text_contains", "params": { "substring": "Dave" } }
[ "multi-step", "read-only", "join" ]
hard
composition
postgres
{"kind": "final_text_contains", "params": {"substring": "Dave"}}
[]
pg_007_count_by_dept
Count rows filtered by joined table
How many employees work in the 'Engineering' department? Give me the count.
[ "postgres" ]
[]
{ "kind": "final_text_regex", "params": { "pattern": "\\b2\\b" } }
[ "multi-step", "read-only", "join" ]
medium
composition
postgres
{"kind": "final_text_regex", "params": {"pattern": "\\b2\\b"}}
[]
pg_008_total_payroll
Aggregate across all rows (sum)
What is the total payroll (sum of every salary) across all employees?
[ "postgres" ]
[]
{ "kind": "final_text_regex", "params": { "pattern": "470,?000" } }
[ "single-tool", "read-only", "aggregate" ]
medium
selection
postgres
{"kind": "final_text_regex", "params": {"pattern": "470,?000"}}
[]
pg_009_list_tables
Enumerate all tables in the schema
What tables exist in this Postgres database? List every one of them.
[ "postgres" ]
[]
{ "kind": "final_text_contains", "params": { "substrings": [ "employees", "departments", "salaries" ], "require_all": true, "case_insensitive": true } }
[ "single-tool", "read-only", "schema" ]
easy
selection
postgres
{"kind": "final_text_contains", "params": {"substrings": ["employees", "departments", "salaries"], "require_all": true, "case_insensitive": true}}
[]
pg_010_recover_table_typo
Error recovery: query a misspelled table name
Query the 'salary' table for the maximum salary amount. (Note: that table name might be misspelled in the prompt — figure out the correct table if so.)
[ "postgres" ]
[]
{ "kind": "final_text_regex", "params": { "pattern": "120,?000" } }
[ "error-recovery", "multi-step", "RQ2" ]
medium
recovery
postgres
{"kind": "final_text_regex", "params": {"pattern": "120,?000"}}
[]
sql_001_count_rows
Count rows in a single table
How many rows are in the 'users' table?
[ "sqlite" ]
[ { "kind": "init_sqlite", "params": { "db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT, age INTEGER); INSERT INTO users (name, age) VALUES ('Alice', 30), ('Bob', 35), ('Carol', 28), ('Dave', 45), ('Eve', 22);" } } ]
{ "kind": "final_text_regex", "params": { "pattern": "\\b5\\b" } }
[ "single-tool", "read-only" ]
easy
selection
sqlite
{"kind": "final_text_regex", "params": {"pattern": "\\b5\\b"}}
[{"kind": "init_sqlite", "params": {"db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT, age INTEGER); INSERT INTO users (name, age) VALUES ('Alice', 30), ('Bob', 35), ('Carol', 28), ('Dave', 45), ('Eve', 22);"}}]
sql_002_filter_count
Count with a WHERE filter
How many users in the 'users' table are older than 30? Tell me the count.
[ "sqlite" ]
[ { "kind": "init_sqlite", "params": { "db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT, age INTEGER); INSERT INTO users (name, age) VALUES ('Alice', 30), ('Bob', 35), ('Carol', 28), ('Dave', 45), ('Eve', 22), ('Frank', 50);" } } ]
{ "kind": "final_text_regex", "params": { "pattern": "\\b3\\b" } }
[ "single-tool", "read-only" ]
easy
selection
sqlite
{"kind": "final_text_regex", "params": {"pattern": "\\b3\\b"}}
[{"kind": "init_sqlite", "params": {"db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT, age INTEGER); INSERT INTO users (name, age) VALUES ('Alice', 30), ('Bob', 35), ('Carol', 28), ('Dave', 45), ('Eve', 22), ('Frank', 50);"}}]
sql_003_describe_schema
List the columns of a table
List every column name in the 'products' table.
[ "sqlite" ]
[ { "kind": "init_sqlite", "params": { "db": "tasks.db", "sql": "CREATE TABLE products (id INTEGER PRIMARY KEY, name TEXT, price REAL, stock INTEGER); INSERT INTO products (name, price, stock) VALUES ('Widget', 9.99, 100);" } } ]
{ "kind": "final_text_contains", "params": { "substrings": [ "id", "name", "price", "stock" ], "require_all": true, "case_insensitive": true } }
[ "single-tool", "read-only", "schema" ]
easy
selection
sqlite
{"kind": "final_text_contains", "params": {"substrings": ["id", "name", "price", "stock"], "require_all": true, "case_insensitive": true}}
[{"kind": "init_sqlite", "params": {"db": "tasks.db", "sql": "CREATE TABLE products (id INTEGER PRIMARY KEY, name TEXT, price REAL, stock INTEGER); INSERT INTO products (name, price, stock) VALUES ('Widget', 9.99, 100);"}}]
sql_004_insert
Insert a new row
Insert a new user into the 'users' table with name 'Alice' and age 42.
[ "sqlite" ]
[ { "kind": "init_sqlite", "params": { "db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT, age INTEGER); INSERT INTO users (name, age) VALUES ('Bob', 35), ('Carol', 28);" } } ]
{ "kind": "sqlite_query_returns", "params": { "db": "tasks.db", "sql": "SELECT COUNT(*) FROM users WHERE name='Alice' AND age=42", "expected_rows": [ [ 1 ] ] } }
[ "single-tool", "write" ]
easy
selection
sqlite
{"kind": "sqlite_query_returns", "params": {"db": "tasks.db", "sql": "SELECT COUNT(*) FROM users WHERE name='Alice' AND age=42", "expected_rows": [[1]]}}
[{"kind": "init_sqlite", "params": {"db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT, age INTEGER); INSERT INTO users (name, age) VALUES ('Bob', 35), ('Carol', 28);"}}]
sql_005_aggregate
Aggregate query (average)
What is the average price of products in the 'products' table? Round to one decimal place.
[ "sqlite" ]
[ { "kind": "init_sqlite", "params": { "db": "tasks.db", "sql": "CREATE TABLE products (id INTEGER PRIMARY KEY, name TEXT, price REAL); INSERT INTO products (name, price) VALUES ('A', 10.0), ('B', 15.0), ('C', 20.0), ('D', 17.0);" } } ]
{ "kind": "final_text_contains", "params": { "substring": "15.5" } }
[ "single-tool", "read-only", "aggregate" ]
medium
selection
sqlite
{"kind": "final_text_contains", "params": {"substring": "15.5"}}
[{"kind": "init_sqlite", "params": {"db": "tasks.db", "sql": "CREATE TABLE products (id INTEGER PRIMARY KEY, name TEXT, price REAL); INSERT INTO products (name, price) VALUES ('A', 10.0), ('B', 15.0), ('C', 20.0), ('D', 17.0);"}}]
sql_006_join
Join two tables and list results
Find the names of every user who has placed at least one order. Use the 'users' and 'orders' tables.
[ "sqlite" ]
[ { "kind": "init_sqlite", "params": { "db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT); CREATE TABLE orders (id INTEGER PRIMARY KEY, user_id INTEGER, total REAL); INSERT INTO users (name) VALUES ('Alice'), ('Bob'), ('Carol'); INSERT INTO orders (user_id, total) VAL...
{ "kind": "final_text_contains", "params": { "substrings": [ "Alice", "Carol" ], "require_all": true } }
[ "multi-step", "read-only", "join" ]
hard
composition
sqlite
{"kind": "final_text_contains", "params": {"substrings": ["Alice", "Carol"], "require_all": true}}
[{"kind": "init_sqlite", "params": {"db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT); CREATE TABLE orders (id INTEGER PRIMARY KEY, user_id INTEGER, total REAL); INSERT INTO users (name) VALUES ('Alice'), ('Bob'), ('Carol'); INSERT INTO orders (user_id, total) VALUES (1, 50.0), (1, 30.0),...
sql_007_list_tables
Enumerate all tables in the database
What tables exist in this database? List every one of them.
[ "sqlite" ]
[ { "kind": "init_sqlite", "params": { "db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER); CREATE TABLE products (id INTEGER); CREATE TABLE orders (id INTEGER);" } } ]
{ "kind": "final_text_contains", "params": { "substrings": [ "users", "products", "orders" ], "require_all": true, "case_insensitive": true } }
[ "single-tool", "read-only", "schema" ]
easy
selection
sqlite
{"kind": "final_text_contains", "params": {"substrings": ["users", "products", "orders"], "require_all": true, "case_insensitive": true}}
[{"kind": "init_sqlite", "params": {"db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER); CREATE TABLE products (id INTEGER); CREATE TABLE orders (id INTEGER);"}}]
sql_008_update
Update an existing row
In the 'users' table, change Bob's age to 50.
[ "sqlite" ]
[ { "kind": "init_sqlite", "params": { "db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT, age INTEGER); INSERT INTO users (name, age) VALUES ('Alice', 30), ('Bob', 35), ('Carol', 28);" } } ]
{ "kind": "sqlite_query_returns", "params": { "db": "tasks.db", "sql": "SELECT age FROM users WHERE name='Bob'", "expected_rows": [ [ 50 ] ] } }
[ "single-tool", "write" ]
easy
selection
sqlite
{"kind": "sqlite_query_returns", "params": {"db": "tasks.db", "sql": "SELECT age FROM users WHERE name='Bob'", "expected_rows": [[50]]}}
[{"kind": "init_sqlite", "params": {"db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT, age INTEGER); INSERT INTO users (name, age) VALUES ('Alice', 30), ('Bob', 35), ('Carol', 28);"}}]
sql_009_delete
Delete a row matching a condition
Delete the user named 'Charlie' from the 'users' table.
[ "sqlite" ]
[ { "kind": "init_sqlite", "params": { "db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT, age INTEGER); INSERT INTO users (name, age) VALUES ('Alice', 30), ('Bob', 35), ('Charlie', 40);" } } ]
{ "kind": "sqlite_query_returns", "params": { "db": "tasks.db", "sql": "SELECT COUNT(*) FROM users WHERE name='Charlie'", "expected_rows": [ [ 0 ] ] } }
[ "single-tool", "write" ]
easy
selection
sqlite
{"kind": "sqlite_query_returns", "params": {"db": "tasks.db", "sql": "SELECT COUNT(*) FROM users WHERE name='Charlie'", "expected_rows": [[0]]}}
[{"kind": "init_sqlite", "params": {"db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT, age INTEGER); INSERT INTO users (name, age) VALUES ('Alice', 30), ('Bob', 35), ('Charlie', 40);"}}]
sql_010_recover_table_typo
Error recovery: query a misspelled table name
Find the maximum age in the 'userz' table and report it. (Note: that name might be misspelled in the prompt — figure out the correct table if so.)
[ "sqlite" ]
[ { "kind": "init_sqlite", "params": { "db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT, age INTEGER); INSERT INTO users (name, age) VALUES ('Alice', 30), ('Bob', 35), ('Carol', 99), ('Dave', 45);" } } ]
{ "kind": "final_text_regex", "params": { "pattern": "\\b99\\b" } }
[ "error-recovery", "multi-step", "RQ2" ]
medium
recovery
sqlite
{"kind": "final_text_regex", "params": {"pattern": "\\b99\\b"}}
[{"kind": "init_sqlite", "params": {"db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT, age INTEGER); INSERT INTO users (name, age) VALUES ('Alice', 30), ('Bob', 35), ('Carol', 99), ('Dave', 45);"}}]
sql_011_oldest_user
Find the row with the maximum value
In the 'users' table, who is the oldest user? Tell me their name.
[ "sqlite" ]
[ { "kind": "init_sqlite", "params": { "db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT, age INTEGER); INSERT INTO users (name, age) VALUES ('Alice', 30), ('Bob', 35), ('Dave', 67), ('Eve', 22);" } } ]
{ "kind": "final_text_contains", "params": { "substring": "Dave" } }
[ "single-tool", "read-only" ]
medium
selection
sqlite
{"kind": "final_text_contains", "params": {"substring": "Dave"}}
[{"kind": "init_sqlite", "params": {"db": "tasks.db", "sql": "CREATE TABLE users (id INTEGER PRIMARY KEY, name TEXT, age INTEGER); INSERT INTO users (name, age) VALUES ('Alice', 30), ('Bob', 35), ('Dave', 67), ('Eve', 22);"}}]
sql_012_create_and_insert
Create a table then insert into it
Create a new table named 'log' with two columns: id (integer) and message (text). Then insert one row whose message is 'hello'.
[ "sqlite" ]
[ { "kind": "init_sqlite", "params": { "db": "tasks.db", "sql": "CREATE TABLE existing (id INTEGER);" } } ]
{ "kind": "sqlite_query_returns", "params": { "db": "tasks.db", "sql": "SELECT message FROM log", "expected_rows": [ [ "hello" ] ] } }
[ "multi-step", "write" ]
hard
composition
sqlite
{"kind": "sqlite_query_returns", "params": {"db": "tasks.db", "sql": "SELECT message FROM log", "expected_rows": [["hello"]]}}
[{"kind": "init_sqlite", "params": {"db": "tasks.db", "sql": "CREATE TABLE existing (id INTEGER);"}}]
sql_013_top_product
Aggregate across a join to find the top item
Two tables exist: 'products' (id, name) and 'sales' (product_id, quantity). Which product has the highest total quantity sold? Tell me its name.
[ "sqlite" ]
[ { "kind": "init_sqlite", "params": { "db": "tasks.db", "sql": "CREATE TABLE products (id INTEGER PRIMARY KEY, name TEXT); CREATE TABLE sales (product_id INTEGER, quantity INTEGER); INSERT INTO products (name) VALUES ('Widget'), ('Gadget'); INSERT INTO sales (product_id, quantity) VALUES (1, 20),...
{ "kind": "final_text_contains", "params": { "substring": "Widget" } }
[ "multi-step", "read-only", "join", "aggregate" ]
hard
composition
sqlite
{"kind": "final_text_contains", "params": {"substring": "Widget"}}
[{"kind": "init_sqlite", "params": {"db": "tasks.db", "sql": "CREATE TABLE products (id INTEGER PRIMARY KEY, name TEXT); CREATE TABLE sales (product_id INTEGER, quantity INTEGER); INSERT INTO products (name) VALUES ('Widget'), ('Gadget'); INSERT INTO sales (product_id, quantity) VALUES (1, 20), (1, 10), (2, 5);"}}]

MCP-Bench tasks (v0.1)

The 70-task dataset behind MCP-Bench — a reliability benchmark for open-source LLM agents using Anthropic's Model Context Protocol (MCP).

At a glance

  • Tasks: 70
  • MCP servers covered:
  • fetch: 13
  • filesystem: 14
  • github: 10
  • memory: 10
  • postgres: 10
  • sqlite: 13
  • Difficulty:
  • medium: 34
  • easy: 24
  • hard: 12
  • Skill (RQ1/RQ2 stratifier):
  • selection: 40
  • composition: 23
  • recovery: 7

How a task is structured

Each task is a JSON object with these fields:

Field Meaning
id Unique task identifier (fs_001_..., sql_003_..., gh_009_..., etc.)
description Short human-readable summary
prompt The exact user message the agent receives
servers MCP server(s) the task needs (["filesystem"], ["postgres"], …)
setup Ordered list of named setup operators to seed the sandbox (write_file, init_sqlite, init_memory)
check A single named check operator that decides pass/fail programmatically
tags Freeform tags (e.g. single-tool, multi-step, RQ2)
difficulty easy / medium / hard
skill selection / composition / recovery / ambiguity

For HF Datasets-friendly loading, setup_json and check_json are also provided as serialized strings of the nested objects.

Named operators

Setup operators (backend/src/mcp_bench/tasks.py):

  • write_file, mkdir, init_sqlite, init_memory

Check operators (every check is deterministic — no LLM judge):

  • file_exists, file_not_exists
  • file_content_equals, file_content_contains
  • final_text_contains, final_text_regex
  • sqlite_query_returns
  • memory_has_entity, memory_not_has_entity, memory_has_relation, memory_entity_observation_contains, memory_entity_count
  • all_of

This lets every pass/fail decision be reproduced from a trace.

Loading

from datasets import load_dataset
ds = load_dataset("keerthanaSubru57/mcp-bench", split="tasks")
print(ds[0]["prompt"])

Citation

@misc{mcp-bench-2026,
  title  = {MCP-Bench: A Reliability Benchmark for Open-Source LLM Agents Using the Model Context Protocol},
  author = {Keerthana S},
  year   = {2026},
  url    = {https://github.com/KeerthanaS04/mcp-bench},
  note   = {v0.1}
}

License

CC-BY 4.0 (this dataset). The harness code is Apache 2.0.

Downloads last month
30

Space using keerthanaSubru57/mcp-bench 1