The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Promise listing page.
Agent-facing MCP server/tools for Promise. The hosted HTTP endpoint is POST /mcp; GET /healthz is available for runtime checks.
This repository is intentionally separate from the Promise platform. It should compose the platform API into agent-friendly tools, not implement canonical backend business logic.
Initial MCP tool surface:
connect_promiseget_todaylist_commitmentsget_commitmentget_commitment_evidenceget_person_contextsearch_memorycapture_notecreate_follow_upmark_resolveddraft_follow_upconnect_promise and unauthenticated tool results are intentionally part of
the product. MCP should help users discover and connect Promise when an agent
needs follow-up memory that Promise can provide.
Unauthenticated tool results include a scoped authorizationUrl that sends the
user to https://app.usepromise.ai/start?source=mcp..., preserves agent
attribution, and uses a relative /mcp/authorize callback inside Promise before
any return to an agent URL.
Example MCP call:
Protected tools return a structured promise_auth_required result when no
Bearer token is present. After the user authorizes /mcp/authorize, Promise
returns a short-lived one-time code to the agent callback. The agent exchanges
that code at /api/mcp/token for a scoped opaque pmcp_... bearer token.
Bearer tool calls are proxied to the Promise platform API, which enforces the
grant's scopes. mark_resolved and draft_follow_up intentionally deep-link
to Promise for user review rather than sending or mutating mail directly.
See docs/ACQUISITION.md for signup flow guidance, docs/SECURITY.md for the
MCP safety boundary, and docs/DEPLOYMENT.md for container/runtime details.