# 6wdb

> A queryable database of AI agents, MCP servers, and developer tools. Every
> tool is summarised in six words and the six Ws (who, what, where, when, why,
> with), and scored with a Proof Score built from verified
> execution telemetry (60%), human stars (25%), and recency (15%).
> Curated entries also carry a seventh W, `watch`: install-risk and provenance
> (credential scope, outbound calls, maintainer type). Null means unassessed.

## Query it (JSON, reads need no auth)
- [Search](/api/v1/search): GET /api/v1/search?q=<text>&kind=<kind>&domain=<domain>&limit=<n>
- [One tool](/api/v1/tools/playwright-mcp): GET /api/v1/tools/{slug}
- [API reference](/api)
- [OpenAPI schema](/openapi.json)
- [Interactive API reference](/docs)

## Report back (feeds Proof Scores)
- [Register a key](/docs#/v1): POST /api/v1/keys {"name": "...", "contact": "..."}
- [Telemetry](/docs#/v1): POST /api/v1/telemetry with X-API-Key header

## Native MCP server
- Endpoint: /mcp (streamable HTTP, stateless). Tools: search_6wdb, get_tool,
  report_telemetry. Point any MCP client at <host>/mcp.

## Publish your own tool (no LLM involved - your words verbatim)
- Add a 6w.json to your repo root (schema: /static/6w.schema.json), then:
  POST /api/v1/submit {"repo_url": "https://github.com/you/your-tool"}

## Bulk snapshot (prefer this over crawling)
- [Full dataset, gzipped JSON](/static/snapshots/6wdb-latest.json.gz) - regenerated nightly.

## For agents
- [Skill file](/static/6wdb-skill.md): how to query 6wdb, with examples.
- Read endpoints are CDN-cacheable (Cache-Control up to 5 min; ETag/304 supported).

## Categories
MCP Server, Agent Framework, Orchestration & Automation, Autonomous Agent,
Coding Assistant, Inference & Serving, Fine-tuning, Vector & Retrieval,
AI Apps & UIs, Observability & Evaluation, Guardrails & Structure