The HTTP API is hosted at https://api.neurynae.com. All endpoints accept JSON; all responses are JSON. Rate-limited, with response caching for repeatable queries.
Endpoints
/v1/searchNatural-language tool search with a clarification loop.
/v1/intelligence/refineDecompose a use-case into structured sub-needs.
/v1/graph/stackBuild a multi-layer tool stack with resolved versions.
/v1/graph/compatibilityVersion-aware compatibility between two tools.
/v1/graph/compareHead-to-head comparison.
/v1/intelligence/verifyValidate agent-suggested tools against the graph.
/v1/intelligence/issueLook up a known GitHub issue (last resort).
Authentication
Every authenticated /v1/* endpoint expects an Authorization: Bearer <token> header. Two ways to get a token:
- Service token — register one for backend or CI usage.
- User token — sign in via the device flow (the MCP server uses this by default).
Full details on Authentication.
Rate limits
The free tier allows 10 req/min and a daily budget that adapts between 100 and 200 based on server load. Pro tier: 60 req/min, 5000/day. See Rate limits.