A dedicated endpoint group for skill, fact, evidence and review operations—the same underlying data the MCP tools expose, reachable from your own backend systems rather than an agent.
Build on context
that can be withdrawn.
Company Brain's data reaches your systems two ways: a REST endpoint layer mounted under a dedicated route group, and the MCP tool surface—already fully documented on this site. Both share the same authorization model: authenticated by API key, scoped to the caller's domain and permission tags on every call.
REST for systems.
MCP for agents.
Thirteen tools, already documented in full: structured lookup, skill search and invocation, evidence retrieval—see the dedicated gateway page for the complete list.
Both surfaces route through the same role and scope checks, so REST and MCP access can never silently diverge on what a caller is allowed to see.
Ten representative routes.
Same data, your own backend.
Mounted under a dedicated route group, not the whole surface — the skill lifecycle, evidence and the authority graph, callable from a server you control instead of an agent.
| Route | What it does | ||
|---|---|---|---|
GET /brain/skills | List published skills the caller's domain grants allow, with confidence and freshness. | ||
GET /brain/skills/{slug} | Full canonical skill document for one published skill. | ||
GET /brain/skills/{slug}/provenance | The cited evidence backing a skill, so a reviewer can check it before it runs. | ||
GET /brain/skills/{slug}/diff | What changed between two versions of the same skill. | ||
POST /brain/skills/{slug}/dry-run | Compile and risk-classify a skill without creating a run. | ||
POST /brain/skills/{slug}/invoke | Execute a published skill through the safe-execution contract; risky steps pause for approval. | ||
GET /brain/runs/{id} | Poll a skill run's status until its human-approval gate clears. | ||
GET /brain/evidence/{id} | Fetch a single cited evidence unit, with a deep link back to its source. | ||
GET /brain/graph/model | The authority graph: which entities exist and how ownership and versions connect them. | ||
GET /brain/tools | Machine-readable list of every MCP tool this gateway exposes, for programmatic discovery. | ||
Internal engineering record, reviewed August 2026. Every route is authenticated by API key and re-checked against the caller's domain and permission-tag scope on each call, matching the MCP surface exactly.
An answer can be flagged.
Not silently kept current.
When a source changes, the affected skill is flagged for review through both surfaces—the same drift signal, whether your integration polls REST or an agent calls MCP.
See the full tool surface →