Missive MCP

An MCP server that connects AI assistants — Claude Desktop and Claude Code — to your Missive team inbox. It gives the assistant 36 tools for contacts, conversations, messages, drafts, posts, shared labels, teams, tasks, and canned responses.

Safe by design. This server can read your inbox, organize it, post internal team comments, and merge conversations — but it cannot send an email or text to anyone outside your team, and it cannot delete anything. missive_create_draft always saves a draft in Missive for a person to review and send; there is no "send" tool and no send parameter.

New here? Follow the friendly step-by-step in SETUP.md — it assumes no prior experience. The rest of this page is the quick reference.

Why it feels turnkey

• The assistant gets usage instructions at connect time. The server ships MCP instructions (the same channel Claude Desktop uses for built-in tools), so the assistant already knows the rules — which list calls need a mailbox filter, that contacts need a contact_book, that drafts never send, and so on.

• Every tool is documented at the call site, grounded in the Missive API docs.

• It documents itself. Ask it to call missive_help (topics: usage, tools, safety, authentication, …) any time.

Related MCP server: IMAP Email MCP Server

Requirements

• Node.js 18+

• A Missive personal access token. In Missive: Preferences → API → Create a new token. (Requires an organization on the Missive Productive plan.)

Install

git clone https://github.com/kanjidoc/missive-mcp.git
cd missive-mcp
npm install
cp .env.example .env        # paste your MISSIVE_API_TOKEN
npm run setup               # validates the token, lists your resource IDs
npm run build

npm run setup confirms the token works and prints your organizations, contact books, teams, and users — handy for filling the optional defaults below — then prints a ready-to-paste launcher config.

Configure your client

The launcher config carries only the start command — never your token (the token lives only in .env). Replace the path with your absolute checkout path.

Claude Desktop — add to claude_desktop_config.json:

{
  "mcpServers": {
    "missive": {
      "command": "node",
      "args": ["/absolute/path/to/missive-mcp/dist/index.js"]
    }
  }
}

Claude Code:

claude mcp add-json missive '{"type":"stdio","command":"node","args":["/absolute/path/to/missive-mcp/dist/index.js"]}'

Restart the client, and the missive tools appear.

Configuration (.env)

Only the token is required; the rest are optional defaults so you don't repeat IDs. Find IDs in Missive → Settings → API → Resource IDs, or run npm run setup.

What's not included (and why)

The Missive API can do a few things this server deliberately leaves out. In plain terms:

• Sending emails or texts to people outside your team. The assistant can write a draft and save it in Missive, but you press send. This is the main safety guardrail — an AI can't fire off a real email on your behalf. (Technically: the drafts endpoint's send / send_at / auto_followup options are not exposed.)

• Deleting things. There is no tool to delete a contact, draft, post, canned response, or label. The only irreversible action offered is merging two conversations, and it's clearly marked as such.

• Analytics reports. Missive can generate inbox/team analytics; those are slow, report-style requests that don't fit a back-and-forth assistant, so they're out.

• Webhooks (real-time event subscriptions). Missive can notify an external app when something happens — that's infrastructure plumbing to set up once, not something an assistant does mid-conversation.

• A few niche send-time options that only matter when actually sending: scheduled send, automated follow-up sequences, and WhatsApp message templates.

Everything else in the Missive REST API — reading and organizing conversations, contacts, tasks, labels, teams, drafts, and internal posts — is available. Any of the above is a small, self-contained addition if you want it later.

Tools (36)

Read-only tools are marked (ro); safely-repeatable updates (idem); irreversible ones (!). See docs/TOOLS.md for the full parameter reference, and call missive_help with topic: "usage" for recipes.

Contacts

Contact books & groups

Conversations

Messages

Drafts & posts

Shared labels

Teams, users & organizations

Responses (canned replies)

Tasks

Help

How the assistant calls it

A few rules the server teaches the model (and worth knowing yourself):

• missive_list_conversations needs a mailbox filter — a boolean like inbox: true, or an id like shared_label / team_inbox. To filter by who's involved, add exactly one of email / domain / contact_organization.

• Contacts need a contact_book; org-scoped writes need an organization (set the MISSIVE_DEFAULT_* vars to avoid repeating them).

• missive_update_contacts replaces the whole infos/memberships array you send — fetch first, then send the full merged array.

• Replying? Pass conversation or references to missive_create_draft and set subject to "Re: …". If you set MISSIVE_DEFAULT_FROM_ADDRESS, the draft uses it automatically; otherwise the assistant takes the from-address from the thread.

Example recipes (the assistant chains these for you):

• Triage: missive_list_conversations(inbox: true) → missive_update_conversations(close: true, …)

• Internal note: missive_create_post(markdown: "…") — posts to the team; no email is sent.

• Draft a reply: missive_get_conversation + missive_list_conversation_messages → missive_create_draft(conversation, to_fields, subject: "Re: …", body)

• Merge duplicates: missive_merge_conversations(source_conversation_id, target_conversation_id)

• Tasks: missive_list_tasks(state: "todo") / missive_create_task / missive_update_task

Rate limits

Missive allows 5 concurrent requests, 300/minute, and 900/15 minutes. The client caps concurrency and automatically retries HTTP 429 honoring Retry-After. For bulk reads, prefer the batch endpoints (e.g. missive_get_message with several ids).

Development

npm run dev     # run with ts-node (no build)
npm run build   # compile to dist/
npm run lint    # eslint
npm test        # vitest
npm run format  # prettier

See CONTRIBUTING.md to add a tool, CLAUDE.md for the architecture and conventions, and SECURITY.md for credential handling. The design rationale lives in docs/superpowers/specs/.

License

MIT © kanjidoc