FollowClose
Sign in

July 6, 2026

Connect FollowClose MCP to AI agents

How to connect FollowClose to Claude Desktop and other MCP clients so your AI agent can generate sequences, manage follow-ups, and export outreach.

If you already use an AI assistant to summarize calls, draft replies, or keep deals moving, FollowClose can give that assistant direct access to your post-call follow-up workflow. Through MCP, your agent can read sales context, generate stored sequences, surface what needs to be sent today, rewrite individual steps, and export sequences into calendar or email-friendly formats without you bouncing between tools.

This matters because post-call follow-up usually breaks in the handoff between conversation and execution. The call happened, the intent is clear, but the actual sequence gets delayed, watered down, or never sent. Connecting FollowClose to an MCP-compatible agent turns that gap into a usable workflow: the agent can take the transcript or notes, create the sequence, adjust one step when needed, and tell you exactly what is due next.

Three concrete use cases

1. Turn a call transcript into a stored sequence

You finish a sales call, drop the transcript into Claude Desktop, and ask it to create a sequence for Maya Chen at Acme Corp. The agent can analyze the call first, confirm the signals are strong enough, then create the sequence in FollowClose and read it back to you with scheduled steps.

2. Run a daily follow-up check

Each morning, your agent can ask FollowClose what is due today in your timezone. Instead of scanning tabs and notes, you can ask a plain-language question like “What follow-ups do I need to send today?” and get back the actual steps, subjects, and body drafts that are ready.

3. Rewrite one weak email without rebuilding everything

If email 3 feels too soft or too long, you do not need to regenerate the whole sequence. Your agent can fetch the stored sequence, update the subject or body directly, or call the single-step regenerate tool with a concrete instruction like “make this shorter and more direct about looping in the VP.”

Quick start

Using an agent that can edit files or run commands, such as Claude Code or Cursor? Paste this link into the chat and ask it to set up FollowClose:

https://followclose.com/mcp.md

That URL serves these same instructions as plain Markdown, written for agents. OAuth-capable clients can connect through the browser consent flow; static keys remain available for clients that need them.

Step 1 — Prefer OAuth

FollowClose publishes MCP OAuth protected-resource metadata at:

https://mcp.followclose.com/.well-known/oauth-protected-resource

An OAuth-capable MCP client can discover the authorization server from that URL, register as a public client, and complete sign-in plus consent with PKCE. This is the preferred connection because you can revoke a named AI app from MCP Access without managing a static secret.

Claude.ai

Open Connectors, add a custom connector using https://mcp.followclose.com, then sign in to FollowClose and approve the requested access. In Claude, enable FollowClose from the Connectors menu before using it in a chat.

Claude Desktop

Open Settings → Connectors, add https://mcp.followclose.com as a custom connector, then complete the FollowClose OAuth prompt before enabling it in a chat.

ChatGPT

Open Settings → Security and login and turn Developer Mode on. Then open Plugins → Browse plugins → +, enter https://mcp.followclose.com, check the confirmation box, and complete the FollowClose OAuth prompt.

Codex

Open Settings → MCP, choose Add server, select Streamable HTTP, and enter https://mcp.followclose.com. Complete the FollowClose OAuth prompt when Codex opens it.

Cursor

Use the one-click installer, then complete the OAuth prompt in Cursor:

If the installer is unavailable, open Settings → Tools & Integrations → MCP, choose Add Custom MCP, and add https://mcp.followclose.com.

VS Code

Use the one-click installer, then complete the same OAuth consent flow:

Step 2 — Use an API key only when OAuth is unavailable

You need an active FollowClose subscription or trial.

  1. Open your dashboard
  2. Go to Settings
  3. Generate a new API key
  4. Copy it immediately — the raw key is shown once

Step 3 — Connect a static-key client

FollowClose uses Streamable HTTP at https://mcp.followclose.com. Choose the configuration for your client.

Claude Code

Run this command and paste the key at the hidden prompt:

read -rs FC_KEY && claude mcp add --transport http followclose https://mcp.followclose.com --header "Authorization: Bearer $FC_KEY"; unset FC_KEY

This is the only setup that invokes a shell command. The key is entered through a hidden prompt rather than placed directly on the command line, but treat your local shell environment accordingly.

Claude Desktop and Cursor

Add this JSON to the relevant config file:

  • Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json on macOS, or %APPDATA%\Claude\claude_desktop_config.json on Windows
  • Cursor: .cursor/mcp.json in your project or user configuration
{
  "mcpServers": {
    "followclose": {
      "url": "https://mcp.followclose.com",
      "headers": { "Authorization": "Bearer fc_live_…" }
    }
  }
}

These JSON clients store the key in their configuration file. Use a separate, revocable key for each client.

VS Code

Add this to your VS Code MCP configuration:

{
  "inputs": [
    { "id": "fc-key", "type": "promptString", "description": "FollowClose API key", "password": true }
  ],
  "servers": {
    "followclose": {
      "type": "http",
      "url": "https://mcp.followclose.com",
      "headers": { "Authorization": "Bearer ${input:fc-key}" }
    }
  }
}

VS Code prompts for the key at runtime, so it does not need to store the secret in plaintext in the configuration file.

Other Streamable HTTP clients

Configure the client with:

Endpoint: https://mcp.followclose.com
Transport: Streamable HTTP
Authorization: Bearer fc_live_…

Step 4 — Verify the connection

Once connected, test it with a simple prompt such as: “Ping FollowClose.”

FAQ

What can my agent actually do with FollowClose MCP?

It can analyze transcripts, create sequences, read stored sequences, list due follow-ups, mark steps sent or skipped, update a stored step, regenerate a single step with AI, and export a sequence as JSON, Markdown, ICS, or ZIP.

Do I need a paid plan for MCP access?

MCP access works on active paid subscriptions and trialing accounts. If billing lapses, existing keys stop working until the subscription returns to active or trialing.

Are API keys reusable after billing is restored?

Yes. FollowClose does not require you to generate a new key after resubscribing. Keys are blocked at verification time when billing is inactive, then resume working once the subscription is active again unless they were revoked or expired.

Can I revoke an OAuth connection?

Yes. Dashboard → Settings → MCP Access lists connected AI apps. Revoking one deletes its opaque access and refresh tokens, stopping that app immediately.

Can I restrict what a key is allowed to do?

Yes. Keys are scope-based. The available scopes are sequence:read, sequence:write, followups:read, and exports:create. A narrower key is the right default when an agent only needs read access or due-followup checks.

What happens if the key is leaked?

Revoke it in Settings and it stops working immediately. FollowClose stores only the hashed key value after creation, not the raw secret.

Does FollowClose log transcript text or generated email content in MCP audit logs?

No. MCP audit logs are metadata-only. They should contain tool names, timing, success or failure, and safe export metadata, not transcript text, generated email bodies, signed download URLs, or prospect identities.

What rate limits should I expect?

Per key, FollowClose currently applies 60/min and 500/day globally, plus tighter AI-tool limits of 10/min and 50/day for sequence creation, context extraction, and step regeneration. Agents should back off instead of retrying in a tight loop.

Can I use it with n8n?

Not as a first-class n8n integration yet. The current requirement is an MCP client that supports Streamable HTTP plus Bearer token auth. If your n8n setup can act as an MCP client with that transport, it can work; otherwise it is still a roadmap item.

Which tool should an agent call first for a raw transcript?

In most cases, extract_followup_context first. It is stateless and lets the agent check whether the transcript has enough usable signal before creating a stored sequence.

What is the safest setup for multiple agents?

Use one key per client or workflow. That keeps revocation clean, keeps last used timestamps meaningful, and reduces the blast radius if one environment is misconfigured.