REST API overview
Every Mneva tool is reachable over HTTPS. The MCP wrapper (mneva-connect) is one client; your own code is another. Use whichever fits.
Base URL
https://mneva.dev/v1
All endpoints are under /v1/. Requests are JSON in, JSON out. UTF-8. Content type application/json on POSTs.
Authentication
One header on every request:
x-mneva-key: mnv_...your_key...
For multi-agent brains, also:
x-mneva-agent: alice
…to attribute observations and (in the future) call-level scoping. See Authentication for key handling and rotation.
Surface map
The 16 MCP tools map to these endpoints. The REST shape is closer to "what the engine does" than to "what the agent says"; some plurals are nouny.
| Method + path | Tool | What it does |
|---|---|---|
POST /v1/remember | remember | record a memory |
POST /v1/recall | recall | semantic search memories |
GET /v1/dreams | recall_dreams | unread overnight thoughts |
POST /v1/believe | believe | assert a belief |
POST /v1/revise | revise | supersede a belief |
POST /v1/evidence | evidence | record SPRT evidence |
GET /v1/evidence/:id | evidence_for | belief audit trail |
POST /v1/flag | flag | record a danger memory |
POST /v1/assess | assess | gut-check + track record |
POST /v1/predict | predict | record a prediction |
POST /v1/resolve | resolve | close a prediction |
GET /v1/predictions/open | predictions_open | unresolved predictions |
GET /v1/calibration | calibration | per-domain track record |
GET /v1/beliefs | — | current beliefs |
GET /v1/state | state | allostatic verdict |
POST /v1/note_about | note_about | multi-agent observation |
GET /v1/outside_views | outside_views | recall observations |
Detail by group:
- Memory endpoints
- Beliefs endpoints
- Predictions endpoints
- Instinct endpoints
- State endpoint
- Multi-agent endpoints
Errors are documented at Errors. All non-2xx return a JSON object with an error string field.