MCP

MCP Server

Connect Claude, Cursor, ChatGPT, and Codex to your Armada workspace.

Overview

Armada ships a hosted MCP server, so an agent can run the same jobs as the product: find people and companies, enrich contact data, save a list, enroll a sequence, work the inbox, and research an account.

The server is a thin layer over the REST API. Every tool calls the same endpoints, with the same authorization, scopes, and rate limits. A tool stops where the REST API stops.

Endpoint

https://mcp.goarmada.co

The server speaks the Streamable HTTP transport. https://app.goarmada.co/api/v1/mcp is the same server.

Authentication

Two ways in. Sign-in is the default for chat apps.

OAuth sign-in. Connect with no API key. The client opens a browser, you sign in, and you allow access for one workspace. The client stores a short-lived mcp_… access token and can refresh it. The connection acts as the person who authorized it. Owners and admins can authorize a connection. Disconnecting the connector in the client leaves your workspace API key alone.

Claude.ai, ChatGPT, and Grok.

API key. A workspace key from Settings → Developers, sent as a Bearer token. The same key the REST API uses. Revoking the key revokes agent access with it.

Authorization: Bearer arm_…

Cursor, Claude Code, Codex, and any client that only accepts a header. Include the word Bearer, then a space, then the key.

Requests without a valid credential are rejected before any tool is listed or run.

Warning

A connected agent can do what you can do in that workspace, including spending credits on enrichment and research, and sending from your connected accounts. Connect agents you trust, and disconnect the ones you no longer use.

Note

Sign-in tokens cannot send SMS, place calls, publish or schedule posts, or enroll social engagers. Enroll, launch, and replies wait for approval in the Armada app. A key from Settings → Developers can send. It still cannot approve its own requests.

Connect a client

Add a custom MCP connector. A browser opens on first connect.

  1. Server URL: https://mcp.goarmada.co
  2. Authentication: Sign in
  3. Allow on app.goarmada.co
  4. In chat, ask the agent to call get_me

Reconnect after a release

A client keeps the tool list it received when it connected, for as long as that connection stays open. New tools and new parameters show up after you reconnect.

The server is stateless, so it does not push a notice when the tool list changes. Reconnect before assuming a documented tool is missing. In Claude Code that is /mcp. Other clients pick it up on reload or restart.

What agents can do

The loop an agent actually needs:

  1. get_me — confirm the workspace
  2. find_people / find_companies — pull a target list
  3. find_email / find_phone / enrich_person / enrich_company — contact data and firmographics. These spend credits.
  4. create_list / append_rows — save the rows
  5. create_sequence / enroll_people — build steps and enroll
  6. launch_campaign — start the campaign. Sign-in connections wait for your approval.
  7. search_conversations / draft_reply / reply_to_lead — work the inbox
  8. research_account / research_dossier / get_research_run — deep research. Poll until the run finishes.
  9. list_playbooks / apply_find_recipe / apply_campaign_playbook — run a playbook

Find and enrich default to provider: auto. Pass a provider slug, or byok:slug for a key saved in Settings → Data Sources.

Seat purchase stays in the Armada app. The REST reference lists the underlying routes.

Costs and limits

  • Enrichment, email and phone find, research, AI copy, and dialer usage bill the same as the product. There is no separate MCP fee.
  • find_email, find_phone, enrich_person, enrich_company, and the research tools spend credits.
  • Creating a list or an empty sequence does not spend credits on its own.
  • Sends use your connected email, LinkedIn, and X accounts, with the same limits as the product.
  • One tool call counts as one REST request against the same rate limits: 180 per minute per key for light reads (900 per workspace), and 40 per minute per key for heavy calls such as enrich, enroll, and sends (160 per workspace). A 429 includes Retry-After.