Contributing to agent-called-what

Thanks for considering it — this project lives or dies by community input.

Ways to help

1. Add a missing MCP server

Edit scripts/lib/mcp_packages.py. Add a tuple to MCP_PACKAGES:

("mcp-server-notion", "pypi", "productivity"),

Format: (package_name, registry, category)

• registry: "npm" | "pypi" | "git"
• category: one of filesystem | database | devops | web | ai | memory | communication | productivity | cloud | social | fintech | observability

Commit convention: feat(data): add mcp-server-notion to seed list

2. Improve the scoring formula

Open an issue first explaining your reasoning. What signal are we under/over-weighting? What real-world behavior does your change better capture? Once discussed, submit a PR against scripts/compose.py.

3. Fix a bug

Check open issues. Look for the good first issue label. Include a failing test or a clear repro in the PR description.

4. Extend the methodology

Got a new signal source? (e.g., PyPI download stats, GitHub Archive BigQuery, MCP.so directory scrape) Open an RFC-style issue. We want the methodology to evolve, but never silently.

Development setup

git clone https://github.com/Zensoro/agent-called-what.git
cd agent-called-what
cp .env.example .env          # paste your GH_PAT
export $(cat .env | xargs)

python scripts/fetch_mcp.py
python scripts/fetch_skills.py
python scripts/compose.py

cd site && npm install && npm run dev   # preview site locally

Rules

• No rm -rf, no --force push, no secrets in commits. The CI checks for .env but don't rely on it.
• No new pip dependencies without discussion. Stdlib only by default.
• Reproducibility > cleverness. If your change makes the pipeline require a new service or key, it needs a really good reason.
• Methodology changes need an issue first. Don't PR a weight tweak without explaining why in an issue.

Code of Conduct

Be kind. Disagree on the weights, not on the person. We're all trying to make AI tooling more legible.