MCP server for AI assistants
Connect Claude, ChatGPT, Cursor, VS Code or n8n to Larvabot over MCP: read-only tools to search, summarise and prioritise your leads and buyer-intent conversations.
Larvabot has an MCP server, so AI assistants like Claude, Cursor and VS Code's Copilot can look through your leads and conversations for you: "which conversations should I answer first today?", "show me the leads that replied this week and what they sell", "draft a follow-up for the café that replied". MCP (Model Context Protocol) is the open standard these assistants use to connect to tools.
The server is read-only. The assistant can search, read and summarise, and it sees the reply and email drafts Larvabot wrote, but it can't post a reply, send an email or change anything. You still do that in Larvabot.
#Server URL and key
https://app.larvabot.com/api/mcpApps that support sign-in (claude.ai, the Claude desktop app, ChatGPT and most newer MCP clients) only need the URL: they open a Larvabot page where you sign in and click Allow access. Everything else uses an API key, the same kind the Zapier and Make apps use: create one under Settings → Integrations → API keys and send it as Authorization: Bearer lb_your_key (or X-API-Key: lb_your_key).
Either way the assistant sees only your own projects. Disconnect a signed-in app, or revoke a key, under Settings → Integrations and it loses access straight away.
#Connect your assistant
#Claude (claude.ai and the desktop app)
- Open Settings → Connectors → Add custom connector.
- Name it Larvabot and paste the server URL.
- Click Connect, sign in to Larvabot if asked, and click Allow access.
On Team and Enterprise plans an owner adds the connector first, under organization settings.
#ChatGPT
- Open Settings → Apps & Connectors → Advanced settings and turn on Developer mode (Plus, Pro, Business and Enterprise plans).
- Click Create, name it Larvabot, paste the server URL and choose OAuth for authentication.
- Sign in to Larvabot when asked and click Allow access. In a chat, pick Larvabot from the tools menu.
#Claude Code
claude mcp add --transport http larvabot https://app.larvabot.com/api/mcp, then run /mcp in Claude Code to sign in. Or with a key:
claude mcp add --transport http larvabot https://app.larvabot.com/api/mcp --header "Authorization: Bearer lb_your_key"#Claude Desktop with an API key (optional)
Prefer a key over the sign-in? Use the small mcp-remote bridge (needs Node.js): open Settings → Developer → Edit Config and add:
{
"mcpServers": {
"larvabot": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://app.larvabot.com/api/mcp", "--header", "Authorization: Bearer lb_your_key"]
}
}
}#Cursor
In Settings → MCP → Add new MCP server, or in ~/.cursor/mcp.json:
{
"mcpServers": {
"larvabot": {
"url": "https://app.larvabot.com/api/mcp",
"headers": { "Authorization": "Bearer lb_your_key" }
}
}
}#VS Code (GitHub Copilot)
In .vscode/mcp.json (or MCP: Add Server from the command palette):
{
"servers": {
"larvabot": {
"type": "http",
"url": "https://app.larvabot.com/api/mcp",
"headers": { "Authorization": "Bearer lb_your_key" }
}
}
}#n8n
Add an MCP Client Tool to an AI Agent, set the endpoint to the server URL, and use Header Auth with Authorization = Bearer lb_your_key.
#Tools
| Tool | What the assistant can do |
|---|---|
pipeline_summary | See what needs you: leads found and replies in the last 7 days, drafts waiting, and the best conversations to answer |
list_projects | List your projects with their counts |
search_conversations | Find posts asking for what you sell, filtered by project, relevance, status, platform or words |
get_conversation | Read one post in full, why it fits, and the reply Larvabot drafted |
search_leads | Find businesses: found, contacted or replied, by fit score or words |
get_lead | Read one lead: why it fits, the detail to mention, the contact, and every email drafted or sent |
#Things to ask
- "What should I reply to today? Give me the top 5 conversations with their drafts."
- "Which leads replied this week? Summarise what each one said yes to."
- "Rewrite the draft reply for conversation 14124 so it's shorter and asks one question."
- "Compare my two projects: where are the warmest leads coming from?"
The assistant reads; you decide. To post a reply or send an email, open the item in Larvabot (every result includes a dashboard_url).