{"id":30956075,"url":"https://github.com/eval-sys/mcpmark","last_synced_at":"2025-09-11T12:07:08.816Z","repository":{"id":309846569,"uuid":"1011688294","full_name":"eval-sys/mcpmark","owner":"eval-sys","description":"MCP Servers are shaping the future of software. MCPMark is a comprehensive, stress-testing benchmark designed to evaluate model and agent capabilities in real-world MCP use.","archived":false,"fork":false,"pushed_at":"2025-09-05T06:24:12.000Z","size":17513,"stargazers_count":117,"open_issues_count":2,"forks_count":4,"subscribers_count":1,"default_branch":"main","last_synced_at":"2025-09-05T08:28:04.736Z","etag":null,"topics":["agentic","benchmark","eval-sys","mcp","mcp-servers","mcpmark","tool-use"],"latest_commit_sha":null,"homepage":"https://mcpmark.ai","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/eval-sys.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":"docs/contributing/make-contribution.md","funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2025-07-01T07:39:50.000Z","updated_at":"2025-09-05T08:10:59.000Z","dependencies_parsed_at":null,"dependency_job_id":"55554a9c-937e-426b-b088-77c5398a7a6a","html_url":"https://github.com/eval-sys/mcpmark","commit_stats":null,"previous_names":["eval-sys/mcpmark"],"tags_count":6,"template":false,"template_full_name":null,"purl":"pkg:github/eval-sys/mcpmark","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/eval-sys%2Fmcpmark","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/eval-sys%2Fmcpmark/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/eval-sys%2Fmcpmark/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/eval-sys%2Fmcpmark/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/eval-sys","download_url":"https://codeload.github.com/eval-sys/mcpmark/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/eval-sys%2Fmcpmark/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":274632631,"owners_count":25321251,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","status":"online","status_checked_at":"2025-09-11T02:00:13.660Z","response_time":74,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["agentic","benchmark","eval-sys","mcp","mcp-servers","mcpmark","tool-use"],"created_at":"2025-09-11T12:03:44.594Z","updated_at":"2025-09-11T12:07:08.806Z","avatar_url":"https://github.com/eval-sys.png","language":"Python","funding_links":[],"categories":["📚 Projects (1974 total)","Python","Agent Harnessing and Evaluation"],"sub_categories":["MCP Servers","Benchmark Reality Check (real-world tool use)"],"readme":"\u003cdiv align=\"center\"\u003e\n\n# MCPMark: Stress-Testing Comprehensive MCP Use\n\n[![Website](https://img.shields.io/badge/Website-mcpmark.ai-4285F4?style=for-the-badge\u0026logo=google-chrome\u0026logoColor=white)](https://mcpmark.ai)\n[![Discord](https://img.shields.io/badge/Join_our_discord-5865F2?style=for-the-badge\u0026logo=discord\u0026logoColor=white)](https://discord.gg/HrKkJAxDnA)\n[![Docs](https://img.shields.io/badge/Docs-000000?style=for-the-badge\u0026logo=mdbook\u0026color=105864)](https://mcpmark.ai/docs)\n[![Hugging Face](https://img.shields.io/badge/Trajectory_Logs-FFD21E?style=for-the-badge\u0026logo=huggingface\u0026logoColor=black)](https://huggingface.co/datasets/Jakumetsu/mcpmark-trajectory-log)\n\n\u003c/div\u003e\n\nAn evaluation suite for agentic models in real MCP tool environments (Notion / GitHub / Filesystem / Postgres / Playwright).\n\nMCPMark provides a reproducible, extensible benchmark for researchers and engineers: one-command tasks, isolated sandboxes, auto-resume for failures, unified metrics, and aggregated reports.\n\n[![MCPMark](https://github.com/user-attachments/assets/dfc06a41-e387-45e3-bc98-db7097ffa3dc)](https://mcpmark.ai)\n\n## What you can do with MCPMark\n\n- **Evaluate real tool usage** across multiple MCP services: `Notion`, `GitHub`, `Filesystem`, `Postgres`, `Playwright`.\n- **Use ready-to-run tasks** covering practical workflows, each with strict automated verification.\n- **Reliable and reproducible**: isolated environments that do not pollute your accounts/data; failed tasks auto-retry and resume.\n- **Unified metrics and aggregation**: single/multi-run (pass@k, avg@k, etc.) with automated results aggregation.\n- **Flexible deployment**: local or Docker; fully validated on macOS and Linux.\n\n---\n\n## Quickstart (5 minutes)\n\n### 1) Clone the repository\n```bash\ngit clone https://github.com/eval-sys/mcpmark.git\ncd mcpmark\n```\n\n### 2) Set environment variables (create `.mcp_env` at repo root)\nOnly set what you need. Add service credentials when running tasks for that service.\n\n```env\n# Example: OpenAI\nOPENAI_BASE_URL=\"https://api.openai.com/v1\"\nOPENAI_API_KEY=\"sk-...\"\n\n# Optional: Notion (only for Notion tasks)\nSOURCE_NOTION_API_KEY=\"your-source-notion-api-key\"\nEVAL_NOTION_API_KEY=\"your-eval-notion-api-key\"\nEVAL_PARENT_PAGE_TITLE=\"MCPMark Eval Hub\"\nPLAYWRIGHT_BROWSER=\"chromium\"   # chromium | firefox\nPLAYWRIGHT_HEADLESS=\"True\"\n\n# Optional: GitHub (only for GitHub tasks)\nGITHUB_TOKENS=\"token1,token2\"   # token pooling for rate limits\nGITHUB_EVAL_ORG=\"your-eval-org\"\n\n# Optional: Postgres (only for Postgres tasks)\nPOSTGRES_HOST=\"localhost\"\nPOSTGRES_PORT=\"5432\"\nPOSTGRES_USERNAME=\"postgres\"\nPOSTGRES_PASSWORD=\"password\"\n```\n\nSee `docs/introduction.md` and the service guides below for more details.\n\n### 3) Install and run a minimal example\n\nLocal (Recommended)\n```bash\npip install -e .\n# If you'll use browser-based tasks, install Playwright browsers first\nplaywright install\n```\n\nDocker\n```bash\n./build-docker.sh\n```\n\nRun a filesystem task (no external accounts required):\n```bash\npython -m pipeline \\\n  --mcp filesystem \\\n  --k 1 \\ # run once to quick start\n  --models gpt-5  \\ # or any model you configured\n  --tasks file_property/size_classification\n```\n\nResults are saved to `./results/{exp_name}/{model}__{mcp}/run-*/...` (e.g., `./results/test-run/gpt-5__filesystem/run-1/...`).\n\n---\n\n## Run your evaluations\n\n### Single run (k=1)\n```bash\n# Run ALL tasks for a service\npython -m pipeline --exp-name exp --mcp notion --tasks all --models MODEL --k 1\n\n# Run a task group\npython -m pipeline --exp-name exp --mcp notion --tasks online_resume --models MODEL --k 1\n\n# Run a specific task\npython -m pipeline --exp-name exp --mcp notion --tasks online_resume/daily_itinerary_overview --models MODEL --k 1\n\n# Evaluate multiple models\npython -m pipeline --exp-name exp --mcp notion --tasks all --models MODEL1,MODEL2,MODEL3 --k 1\n```\n\n### Multiple runs (k\u003e1) for pass@k\n```bash\n# Run k=4 to compute stability metrics (requires --exp-name to aggregate final results)\npython -m pipeline --exp-name exp --mcp notion --tasks all --models MODEL\n\n# Aggregate results (pass@1 / pass@k / pass^k / avg@k)\npython -m src.aggregators.aggregate_results --exp-name exp\n```\n\n### Run with Docker\n```bash\n# Run all tasks for a service\n./run-task.sh --mcp notion --models MODEL --exp-name exp --tasks all\n\n# Cross-service benchmark\n./run-benchmark.sh --models MODEL --exp-name exp --docker\n```\n\nPlease visit `docs/introduction.md` for choices of *MODEL*.\n\nTip: MCPMark supports **auto-resume**. When re-running, only unfinished tasks will execute. Failures matching our retryable patterns (see [RETRYABLE_PATTERNS](src/errors.py)) are retried automatically. Models may emit different error strings—if you encounter a new resumable error, please open a PR or issue.\n\n---\n\n## Service setup and authentication\n\n| Service     | Setup summary                                                                                                  | Docs                                  |\n|-------------|-----------------------------------------------------------------------------------------------------------------|---------------------------------------|\n| Notion      | Environment isolation (Source Hub / Eval Hub), integration creation and grants, browser login verification.     | [Guide](docs/mcp/notion.md)           |\n| GitHub      | Multi-account token pooling recommended; import pre-exported repo state if needed.                              | [Guide](docs/mcp/github.md)           |\n| Postgres    | Start via Docker and import sample databases.                                                                   | [Setup](docs/mcp/postgres.md)         |\n| Playwright  | Install browsers before first run; defaults to `chromium`.                                                      | [Setup](docs/mcp/playwright.md)       |\n| Filesystem  | Zero-configuration, run directly.                                                                               | [Config](docs/mcp/filesystem.md)      |\n\nYou can also follow [Quickstart](docs/quickstart.md) for the shortest end-to-end path.\n\n---\n\n## Results and metrics\n\n- Results are organized under `./results/{exp_name}/{model}__{mcp}/run-*/` (JSON + CSV per task).\n- Generate a summary with:\n```bash\n# Basic usage\npython -m src.aggregators.aggregate_results --exp-name exp\n\n# For k-run experiments with single-run models\npython -m src.aggregators.aggregate_results --exp-name exp --k 4 --single-run-models claude-opus-4-1\n```\n- Only models with complete results across all tasks and runs are included in the final summary.\n- Includes multi-run metrics (pass@k, pass^k) for stability comparisons when k \u003e 1.\n\n---\n\n## Model and Tasks\n- **Model support**: MCPMark calls models via LiteLLM — see the LiteLLM docs: [`LiteLLM Doc`](https://docs.litellm.ai/docs/). For Anthropic (Claude) extended thinking mode (enabled via `--reasoning-effort`), we use Anthropic’s native API.\n- See `docs/introduction.md` for details and configuration of supported models in MCPMark.\n- To add a new model, edit `src/model_config.py`. Before adding, check LiteLLM supported models/providers. See [`LiteLLM Doc`](https://docs.litellm.ai/docs/).\n- Task design principles in `docs/datasets/task.md`. Each task ships with an automated `verify.py` for objective, reproducible evaluation, see `docs/task.md` for details.\n\n---\n\n## Contributing\n\nContributions are welcome:\n1. Add a new task under `tasks/\u003ccategory_id\u003e/\u003ctask_id\u003e/` with `meta.json`, `description.md` and `verify.py`.\n2. Ensure local checks pass and open a PR.\n3. See `docs/contributing/make-contribution.md`.\n\n---\n\n## Citation\n\nIf you find our works useful for your research, please consider citing:\n\n```bibtex\n@misc{mcpmark_2025,\n  title        = {MCPMark: Stress-Testing Comprehensive MCP Use},\n  author       = {The MCPMark Team},\n  howpublished = {\\url{https://github.com/eval-sys/mcpmark}},\n  year         = {2025}\n}\n```\n\n## License\n\nThis project is licensed under the Apache License 2.0 — see `LICENSE`.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Feval-sys%2Fmcpmark","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Feval-sys%2Fmcpmark","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Feval-sys%2Fmcpmark/lists"}