mcp-observatory AGENTS.md

Agent Guide For MCP Observatory

MCP Observatory helps agents test other MCP tools before relying on them. Keep changes focused on trust, compatibility, reports, and real MCP target evidence.

High-Leverage Agent Tasks

• Add a real MCP server target that can run in CI without secrets.
• Follow docs/target-contribution-guide.md to add one Safety Index target and generated evidence.
• Use docs/agent-tasks.md for copy-pasteable bot and maintainer contribution tasks.
• Use docs/launch.md and docs/code-scanning-demo.md when explaining the project to maintainers.
• Improve the Markdown or JSON report so autonomous agents can decide whether a tool is safe to use.
• Make startup failure diagnosis clearer.
• Improve setup-ci --all so maintainers can add read-only MCP checks with less friction.
• Add fixtures that prove tool, prompt, resource, schema drift, or security behavior.

Avoid

• generic dashboards
• broad adapter rewrites without a concrete failing target
• marketing-only changes
• CI that requires private credentials by default
• noisy blocking checks before the signal is useful

Useful Commands

npm install
npm run lint
npm run typecheck
npm test
npm run smoke

For a real target sweep:

npm run integration:real

For target-registry contributions:

npm run safety-index
npm test -- tests/safety-index.test.ts

For campaign-attributed launch work:

npx @kryptosai/mcp-observatory test <safe startup command> --campaign agent-ci
npx @kryptosai/mcp-observatory setup-ci --all --command "<safe startup command>" --sarif --campaign agent-ci

Bot Maintainer Contribution Path

If you maintain an autonomous agent, OpenClaw-style productivity system, MCP gateway, or coding-agent runtime, the best contribution is a safe target:

Project:
Safe startup command:
Needs secrets? yes/no
Agent tools/skills that depend on MCP:
What should the agent know before trusting this tool?

Use docs/agent-runtime-quickstart.md for the detailed path.