MCP uses the same public contract and authorization rules as other clients. Approve the smallest scope, inspect the advertised tools, and require human confirmation before consequential writes.
Current connection model
| Element | Current behavior |
|---|---|
| Endpoint | https://app.mailsequence.com/api/mcp |
| Transport | Stateless HTTP JSON-RPC request and response; no streaming |
| Protocol | MCP protocol version 2025-06-18 |
| Identity | One active workspace selected during the consent flow |
| Authorization | Approved workspace scopes; provisioning keys are refused |
| Rate limit | The same per-key limit and Retry-After behavior as the API |
1. Create a scoped connection
Begin the MailSequence consent flow from a supported MCP client or connection link while signed in to the intended workspace. Review the client label, connection name, and requested scopes. Only workspace owners and admins can approve the connection.
The callback returns the connection details, endpoint, and bearer token once. Save the token in the client's secure credential store. The consent link is temporary and single-use. It does not provide general OAuth client registration or refresh tokens.
2. Understand the tool surface
| Scope | Available tools |
|---|---|
contacts:read | list_contacts, get_contact |
contacts:write | create_contact, update_contact |
sequences:read | list_sequences |
campaigns:read | list_campaigns |
enrollments:write | enroll_contact |
imports:write | import_contacts |
The server filters tools/list to the connection's scopes and checks authorization again when a tool is called. MCP currently exposes eight tools. The wider REST API remains separate.
3. Use a reviewable agent workflow
- Ask the client to list campaigns or contacts before selecting IDs.
- Require it to summarize the proposed audience, destination, and conflict policy.
- Review a small import or contact update before authorizing broader work.
- Require explicit confirmation before
enroll_contact. - Record returned IDs and batch handles so the operation can be reconciled.
Tool descriptions do not replace operator intent. An agent with write scope can perform the corresponding write, so put confirmation and source-data review in the client workflow.
4. Preserve API safety rules
- Contact creation still deduplicates inside the workspace and never un-suppresses.
- Import still enforces field, batch, conflict-policy, and plan-cap rules.
- Enrollment still checks contact, campaign, state, and tenant boundaries.
- Mutating calls can forward an idempotency key for safe retries.
- A tool cannot call a resource outside its approved scopes.
5. Revoke access when the client no longer needs it
The current product does not provide a dedicated UI for listing and revoking MCP connections. The practical kill switch is to revoke the backing workspace API key from the Developer tab. That invalidates both MCP and REST use of the token.
Sources and product basis
- Model Context Protocol: Authorization
Authorization concepts for protected remote MCP servers. - Model Context Protocol: Transports
HTTP transport behavior and security considerations. - MailSequence public API reference
The underlying workspace-scoped contract used by MCP tools. - MailSequence developer platform
Current MCP, API, scope, tenancy, and plan-access claims.