Repclose

Read your Repclose account plans from Claude, ChatGPT or Cursor: list and search accounts, open one account's plan with its dated sources, and see the people at each account. Read-only. Needs a Repclose account and API key.

Servidor MCP hospedado

npx add-mcp 'https://repclose.com/mcp'

Instala no Claude Code, Codex, Cursor e outros

Documentação

Connect Claude, ChatGPT or Cursor to your Repclose accounts. Ask "Which accounts should I call this week?", "What changed at Chewy since last week?" or "I'm talking to Satish and José at Chewy. Who am I missing?" in your own chat. Your AI only reads: it can't edit a plan, save a note or start research. Use Repclose's MCP alongside your CRM's MCP; Repclose does not connect to your CRM.

1. Make a key

Sign in, open Connect your AI, and create a key. Copy it then; it's shown once. Revoke it there any time.

Server address: https://repclose.com/mcp

2. Add it to your AI

Claude Code

claude mcp add --transport http repclose https://repclose.com/mcp \
  --header "Authorization: Bearer YOUR_KEY"

Claude Desktop

Open Settings, Developer, Edit Config. Add this to claude_desktop_config.json and restart Claude Desktop. It needs Node.js for the mcp-remote bridge.

{
  "mcpServers": {
    "repclose": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://repclose.com/mcp",
               "--header", "Authorization:${AUTH}"],
      "env": { "AUTH": "Bearer YOUR_KEY" }
    }
  }
}

Cursor

Open Settings, MCP, Add new server (or edit ~/.cursor/mcp.json):

{
  "mcpServers": {
    "repclose": {
      "url": "https://repclose.com/mcp",
      "headers": { "Authorization": "Bearer YOUR_KEY" }
    }
  }
}

ChatGPT

Turn on developer mode (Settings, Connectors, Advanced), then add a custom connector with the server address above. Choose the option that sends an access token or API key, and paste your key as the Bearer token. If your plan only offers OAuth sign-in, ChatGPT can't connect yet.

What your AI can do

ToolWhat it returns
list_accountsYour accounts, best fit first when a fit score exists.
search_accountsAccounts matching a name, domain, plan text or person.
get_account_planOne account's plan, with sources and their publication dates where known. By default a short answer: At a glance and the names of the other sections. Ask for sections by name, or for the whole plan. It says when no research has run and lists what research hasn't found yet, never as findings.
get_peopleThe people at one account, the roles they play, and one line on why each cares, with its source. It says why a title is empty.
who_is_missingGive the names already in the deal, as a list of name strings. Returns the people who matter and aren't in it, best buying role and fit first, and for each name you gave whether Repclose shows them as current, left, changed title, unverified or not found. Names match exactly, ignoring case and accents; nothing is guessed.
what_changedThe newest research refresh's findings that matter, with sources and dates. Add a day to get every research refresh after it, oldest first; if there are too many refreshes, it says how many older ones it left out.
who_to_approachWho to approach this week, best first, with reasons and sources. Uses Repclose's own ranking; no new research.
get_personThe deeper look at one person: what their person research found, with sources and dates. If nobody has researched them, it says so; start that from their card in Repclose.

Accounts research hasn't scored yet have a null fit_score. Each key is limited to 20 calls a minute and 200 a day, shared across all web workers; a search counts as 3 calls. Past the limit, your AI gets an error that says how many seconds to wait. Made-up keys without the real key shape are rejected without a Firestore read. Unknown keys with the real shape are limited to 20 tries a minute per network address in each worker. Your private notes, account notes and your own plan items are never shared with your AI; the plan and people answers say how many plan items were held back. People who opted out of Repclose are left out.