How to connect Claude or ChatGPT to Coolie with MCP

Claude (web, desktop and mobile)

1. 1

   Open Claude's connectors

   In Claude's settings, open Connectors and click Add custom connector. On a Team or Enterprise plan, an Owner adds it in Organization settings → Connectors.

2. 2

   Paste Coolie's address

   Enter https://trycoolie.ai/api/mcp, keep Sign in now, click Add, then Connect.

3. 3

   Allow the connection

   Log in to Coolie and click Allow. The connection gets a key of its own, listed in Settings → API keys, where you can revoke it.

ChatGPT

1. 1

   Turn on developer mode

   In ChatGPT's settings, open Connectors, with developer mode on.

2. 2

   Create the connector

   Create a connector, paste https://trycoolie.ai/api/mcp and choose OAuth.

3. 3

   Allow the connection

   Log in to Coolie when ChatGPT asks, and click Allow.

Claude Code

One command adds the server; then type /mcp in Claude Code to sign in.

claude mcp add --transport http coolie https://trycoolie.ai/api/mcp

claude mcp add --transport http coolie https://trycoolie.ai/api/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

Cursor, VS Code and other clients

In Coolie, Settings → MCP has a button for each: Cursor and VS Code open with Coolie added, and sign in on first use. Any other client takes the address as a remote MCP server over Streamable HTTP, with the OAuth sign-in, or an API key in an Authorization: Bearer header or an x-api-key header.

Call the server without an assistant

The server speaks JSON-RPC 2.0 over HTTP POST, without a session: one request, one JSON answer. tools/list lists the tools, tools/call runs one. A check with curl, then in Python:

curl https://trycoolie.ai/api/mcp \
  -H "Authorization: Bearer $COOLIE_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"list_leads","arguments":{"limit":5}}}'

import os

import requests

response = requests.post(
    "https://trycoolie.ai/api/mcp",
    headers={
        "Authorization": f"Bearer {os.environ['COOLIE_API_KEY']}",
        "Accept": "application/json, text/event-stream",
    },
    json={"jsonrpc": "2.0", "id": 1, "method": "tools/list"},
    timeout=60,
)
for tool in response.json()["result"]["tools"]:
    print(tool["name"], "-", tool["description"])

What the assistant can do

Each tool is one call of the API, with its rules and its costs:

• Account and catalogs: get_account, list_signals, get_icp_options, search_catalog.
• Agents: list_agents, get_agent, create_agent, update_agent, start_agent, pause_agent, stop_agent, delete_agent.
• Leads: list_leads, find_contact, research_leads, delete_leads, rate_lead, add_lead_insight, remove_lead_insight, write_lead_sequence.
• Campaigns: get_campaign, update_campaign, launch_campaign, pause_campaign, write_step, list_queue, add_to_queue, update_queue, get_queue_item, edit_queue_messages, regenerate_message.
• Settings: get_settings, update_settings.

The server tells the assistant to confirm with you before it deletes an agent, which deletes its leads, and before it launches a campaign, which sends real messages from your accounts.

Limits and costs

• Each message counts as one request of your account's 60 a minute.
• A tool costs what the same action costs in the app: reading is free; an email found costs 0.1 credit, a phone 1, a lead researched 1.
• The connection acts as you: it sees your agents and leads, and nobody else's.

What a tool call answers

The tool's result is the API's JSON, as text for the model; isError is true when the call failed, with the API's error code and message in the text. The values are examples.

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "{\"leads\":[{\"runId\":\"ui-5b9e2c41\",\"prospectId\":\"p-41b7c2d9e0aa\", …}],\"nextCursor\":null}"
      }
    ],
    "isError": false
  }
}