Skip to main content
Komo ships a hosted MCP server so your AI assistant can run your prospecting directly from chat: search 200M+ people and companies, manage your CRM, build lists, run signal agents, and launch LinkedIn + email outreach — using your existing Komo plan and credits.

Server

Endpointhttps://mcp.komo.ai/mcp
TransportStreamable HTTP (stateless)
AuthKomo API key (pk_…:sk_…)
Tools42, across 8 categories
Promptsprospect, daily_briefing, campaign_review (shown in your client’s prompt picker / as slash commands)
CostIncluded with your plan — actions draw from the same credit balance as the app

Set up your client

Step-by-step setup for Claude, Claude Code, Cursor, and Codex — or create a key and get pre-filled snippets at app.komo.ai/connect.

Authentication

Create an API key at app.komo.ai/settings/api-keys. The key is a public/secret pair (pk_…:sk_…) shown once at creation. Send it any of three ways:
Authorization: Bearer pk_xxx:sk_xxx     # most MCP clients
x-api-key: pk_xxx:sk_xxx                # alternative header
https://mcp.komo.ai/mcp/pk_xxx:sk_xxx   # key-in-URL, for clients that can't set headers
Access is scoped to the workspace that owns the key. Revoking the key at any time disconnects every client using it.

Your guardrails travel with you

Every tool call goes through the same server-side enforcement as the Komo app — an AI client can never do more than your account can:
  • Credits & plan. Credit-consuming tools (search, enrichment, sends) check your balance first and fail with a clear top-up message when it’s insufficient. Paid-plan gates (e.g. starting a campaign, connecting a LinkedIn seat) apply unchanged.
  • LinkedIn safety caps. Standalone sends and campaign sends draw from one shared per-seat budget: daily and weekly invitation/message limits, plus pacing between actions. When a cap or the pacing window is hit, the tool returns a retry time instead of sending.
  • Review mode. Campaigns with review mode on still hold every recipient for human approval, whichever surface enrolled them.
Tools that send real outreach (start_campaign, send_linkedin_invitation, send_linkedin_message, engage_linkedin_post) are annotated as destructive and instruct the assistant to confirm with you first. In Claude, we recommend keeping these on “ask before running”.

What you can ask

  • “Find 25 VPs of Sales at Series B fintech companies in New York and save them to a new list called ‘Q3 fintech’.”
  • “What’s in my CRM for Acme Corp? Summarize recent activity before my call.”
  • “Run my signal agent and tell me who engaged with competitor posts this week.”
  • “Draft a 4-step LinkedIn + email campaign for the ‘Q3 fintech’ list — review mode on.”
  • “Check my LinkedIn inbox for replies and summarize them.”
  • “How many credits do I have left?”

For agents

A machine-readable overview lives at https://api.komo.ai/llms.txt, and the public REST surface (the same endpoints the MCP tools wrap) is described by https://api.komo.ai/api/public-openapi.json.