Odel
Promise

Promise

@usepromiseTypeScriptUpdated 4 days ago

Evidence-backed follow-through memory for what you owe, await, and need to remember.

View on GitHubSign in to install
Server endpointStreamable HTTPNo authProbed
https://mcp.usepromise.ai/mcp

This is the third-party server itself — Odel doesn't run it. Hitting this URL directly talks straight to the upstream server with no auth or proxying. Connect through Odel to front it with managed auth.

Promise MCP

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.

Boundary

  • Call the Promise platform API for data access and mutations.
  • Do not access Postgres, Redis, queues, provider tokens, or migrations directly.
  • Keep authorization, provenance, retention, provider behavior, and canonical domain rules in the platform.
  • Consume explicit platform contracts instead of duplicating procedure or capability names by hand.

Initial MCP tool surface:

  • connect_promise
  • get_today
  • list_commitments
  • get_commitment
  • get_commitment_evidence
  • get_person_context
  • search_memory
  • capture_note
  • create_follow_up
  • mark_resolved
  • draft_follow_up

connect_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.

Development

npm install
npm run build
npm run smoke
npm run tools
npm run start

Example MCP call:

curl -sS http://localhost:3000/mcp \
  -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

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.