# toolidx

> Independent verification and directory service for MCP servers and AI tools. Machine-readable status, evaluation scores, and structured metadata.

toolidx indexes and verifies MCP (Model Context Protocol) servers. Each listing includes a verified description, install command, tool schemas from live QC testing, and multi-model evaluation scores. All data is queryable via REST API.

## Quick Start for Agents

1. GET /v1/servers?status=active&qc_status=passed — list verified, installable servers
2. GET /v1/servers/{id} — full record including tool schemas and scores
3. Use the `install_command` field to install the server
4. Check `tool_schemas` to understand available tools before integrating

## Endpoints

- [GET /v1/status](https://toolidx.dev/v1/status): Service health and index metadata
- [GET /v1/servers](https://toolidx.dev/v1/servers): List servers — filter by status, qc_status; paginate with `?limit=N&page=N` (limit max 100, page is 1-indexed; `offset` is ignored)
- [GET /v1/servers/{id}](https://toolidx.dev/v1/servers/{id}): Full server record
- [GET /openapi.json](https://toolidx.dev/openapi.json): OpenAPI 3.1 specification

## Response Envelope

Every response includes `last_updated` (ISO 8601 UTC) at the top level so agents can assess data freshness without a separate status call.

## Server Record Fields

- `id` — stable slug from repository URL (e.g. github-com-owner-repo-name)
- `name` — display name
- `description` — AI-generated, quality-gated description
- `install_command` — verified install command
- `package_type` — npm | uvx | pip
- `tool_schemas` — array of tool definitions from live QC install test
- `tool_count` — number of tools the server exposes
- `qc_status` — pending | passed | failed | error | skipped
- `quality_score` — weighted eval score 0–10 (Accuracy 30%, Specificity 25%, Actionability 20%, Trust 15%, Completeness 10%)
- `sanity_score` — multi-model consensus score 0–10
- `status` — active | pending | rejected

## Verification Pipeline

Each server passes through:
1. AI description generation (multi-model)
2. Sanity Panel cross-review (5 independent agents, divergence flagged)
3. Quality gate (Claude API cold reviewer — approve / revise / reject)
4. QC install test (live npm/uvx/pip install, tool schema introspection)

## Index Stats

- 2,053 servers indexed
- Last updated: 2026-05-08 05:29:58

## More

- API docs: https://toolidx.dev/docs
- OpenAPI spec: https://toolidx.dev/openapi.json
- Related: https://agenticwatch.dev