Integration & API reference

This page describes the surfaces intended for outside integration. It is deliberately scoped.

Ceiling — Most of the backend speaks tRPC and is internal plumbing: it can change between releases and is not a published, stable API. Don't build integrations against it. The surfaces below are the ones designed to be spoken to from outside.

MCP server#

A stdio-based MCP server exposes an agent's identity, memory, and governance context to MCP clients (Claude Code, Codex, Cursor). Configure the client to launch it; auth tokens are managed in Settings → MCP. This is the way to give an external editor-agent the same identity and memory as a damn.dev agent.

Connector ingestion endpoints#

Outside workers report activity into your audit log through token-authenticated HTTP endpoints. You don't usually call these by hand — the Connect panels generate the exact config — but they are the stable integration contract:

Surface	Purpose
Claude Code	A hook reports tool use; an optional PreToolUse gate returns an allow/deny decision.
Hermes / MCP	An interposition proxy submits each MCP call for an allow/ask/deny decision before forwarding.
OpenTelemetry	A standard OTLP endpoint accepts logs/spans from any OTLP emitter (observe-only).
External worker events	A generic per-worker token ingests declared events.

Each connector token is workspace-scoped and minted by an operator. The honest enforcement label for each (declared / observed / gated) is in Governing AI you don't run.

Version & update#

• GET /api/version returns the running version plus update-check telemetry.
• GET /version.json (on the public site) is what instances poll to detect updates.

Getting the exact snippets#

Open Oversight → Registry → Connect for the copy-paste config for any connector, built against your instance's URL. See MCP server & connectors for the walkthrough.

Next#

• Governing AI you don't run — what each endpoint enforces vs. observes.
• MCP server & connectors — setup walkthrough.