Claude Code MCP setup for GTM
Quick answer
Claude Code connects to MCP servers with one command, claude mcp add, and signs in to hosted servers with OAuth. Our GTM sessions run with our CRM, LinkedIn outreach, Slack and our contact database connected, so Claude reads the real state of a deal before it acts. The setup is quick; the traps are in signing in, in slow servers and in each tool's own limits.
- Published by
- Anthropic. Claude Code is the client; each server comes from its tool
- How it runs
- From the command line, with claude mcp add
- Get it
- code.claude.com/docs/en/mcp
What your assistant can do with it
- Work in your CRM: read a deal, write the note, set the next task (Attio, HubSpot)
- Run LinkedIn outreach: read replies, stop a lead, build a campaign (HeyReach)
- Search and enrich contacts, check credits before spending (Apollo, Clay)
- Read what the team decided before acting, and post when asked (Slack)
How to connect it
- Add a hosted server for your user: claude mcp add --transport http -s user <name> <url>.
- Sign in: in an interactive Claude Code session, run /mcp; or run claude mcp login <name> in a real terminal, then approve the access in the browser page it opens.
- Check it: claude mcp list should show the server as connected.
- Open a new session: a server added during a session is only loaded by sessions started afterwards.
- Test each server with a real read (a record, a campaign, a credit balance) before you rely on it.
How we use it
Every working session we open on Enjyn's sales starts with the same servers: Attio for the CRM, HeyReach for LinkedIn, Slack for what the team decided, Apollo for contacts. Claude reads the deal, the conversation and the channel before it writes anything, and a rules file tells it what it may do alone and what needs our OK, like sending a message.
Where it breaks
- The sign-in command needs a real terminal: launched from a script it stops with an error about stdin not being a terminal.
- Some servers are slow to start. Our self-hosted n8n server took 20 to 35 seconds, more than the default wait; raising the MCP_TIMEOUT environment variable fixed it.
- A server can list its tools and still refuse every call: HeyReach answers the tool list with any key and returns 401 on real calls without the right one.
- Each tool keeps its own limits behind the server: a monthly credit cap per MCP user in Clay, paid company searches in Apollo, missing writes in Attio. Read each server's page before automating with it.
- OAuth can grant more than you need. Check the scopes on the consent page before you approve.
Our take. Worth the twenty minutes: once the CRM, LinkedIn and Slack are connected, Claude works from the real state of each deal instead of what you paste in. Add servers one at a time, test each with a real call, and write down what it may do without asking.
Frequently asked questions
How do I add an MCP server to Claude Code?
Run claude mcp add --transport http <name> <url> for a hosted server, then sign in with /mcp in a Claude Code session. Use -s user to make it available in every project.
Why doesn't my new MCP server show up?
A server added during a session is loaded by the sessions you open afterwards. Start a new session, and check claude mcp list.
Which MCP servers are worth connecting for sales?
Your CRM first, then the tool that sends (LinkedIn or email), then your contact database. Our MCP servers page covers the ones we run, with their limits.
More servers we run: MCP servers for GTM.
Want this wired into your stack
We build and run GTM systems where Claude works across your CRM, data and sending tools, with rules it has to follow.
Book a GTM audit