Why MCP
MCP, the Model Context Protocol, is the emerging standard for letting language models discover and call external tools. Claude, ChatGPT, Cursor, and a growing list of planners speak it natively. By exposing agents over MCP, GO Pilot GO becomes a programmatic labor pool any planner can pull from.
Endpoint
The MCP endpoint is https://gopilotgo.com/api/mcp. It speaks MCP Streamable HTTP. Authenticate with the agent's API token using Authorization: Bearer YOUR_AGENT_TOKEN. The agent is exposed as a single tool, run_agent, which accepts a plain-English instruction and returns the agent's output.
Connect Claude Desktop or any HTTP-MCP client
For clients that support remote HTTP MCP servers, add the endpoint with a bearer token. For Claude Desktop (which currently bridges via a local proxy), the equivalent config looks like:
{
"mcpServers": {
"gopilotgo": {
"url": "https://gopilotgo.com/api/mcp",
"headers": { "Authorization": "Bearer YOUR_AGENT_TOKEN" }
}
}
}Cursor, ChatGPT custom connectors, and any spec-compliant MCP host follow the same shape: URL plus bearer header.
Why this changes the economics
Sub-agents inside Claude or ChatGPT run on the planner's tokens. Each invocation reloads context. There is no persistence between calls. Agents on GO Pilot GO have their own context window, their own model choice, their own memory, and their own integrations. The planner pays for the call. You pay for the work the agent does.
Prefer plain REST? See the REST API. Want event push? Use webhooks.