MCP
Promise should be useful wherever your agents work.
Promise MCP is the planned agent interface for what you owe, what you are waiting on, and what matters about people.
The point is not mailbox access
Generic email MCP servers can search or send mail. Promise MCP should do something more specific: expose the follow-up memory Promise has already derived from email and quick notes, with the original context close by when available.
Initial tool surface
get_today: Return what currently deserves the user's attention.list_commitments: List what the user owes and what the user is waiting on.get_commitment_evidence: Show why Promise believes a follow-up exists.get_person_context: Summarize what matters before the user speaks with someone.search_memory: Search derived Promise memory with user authorization.capture_note: Let a user tell Promise something to remember.draft_follow_up: Create a draft or deep link for review, never send directly.
Signup path through MCP
Yes: Promise should treat MCP as an acquisition surface. If an agent cannot answer "what did I promise Sarah?" because the user has not connected Promise, the MCP should return a clear onboarding result with a Promise signup URL and a short explanation of why connecting Gmail or Outlook enables the answer.
The first safe version should drive users to Promise to connect their account, then later return them to the agent after authorization. It should not ask the agent to collect credentials.
Safety contract
- Call the Promise platform API; do not access databases directly.
- Use explicit OAuth scopes and let users revoke agent access.
- Return evidence and confidence, not unsupported certainty.
- Draft follow-ups, but do not send email directly in v1.
- Deep-link to Promise for confirmation when an action is sensitive.