MCP

Model Context Protocol

Connect ChatGPT, Codex, Claude, or any Streamable HTTP MCP client to Stalkr’s complete tool catalog.

Server URL

https://www.stalkr.ai/mcp

The endpoint supports current per-request MCP and the stateless 2025 Streamable HTTP flow. Tools include read/write/destructive/open-world annotations so compatible clients can present appropriate confirmations.

ChatGPT setup

Add a custom MCP connection using the server URL above.
ChatGPT discovers Stalkr’s OAuth endpoints from the protected-resource metadata.
Sign in to Stalkr and review the requested scopes.
Approve access. The new connection appears in Dashboard → Developer → Connections.

No API key pasted into ChatGPT

ChatGPT uses OAuth rather than a custom API-key header. Access tokens last one hour; rotating refresh tokens keep the approved connection active until you revoke it.

Other MCP clients

OAuth-capable clients should use discovery. Trusted server-side clients that support custom headers may use a scoped Stalkr API key.

Generic configuration

{
  "mcpServers": {
    "stalkr": {
      "url": "https://www.stalkr.ai/mcp",
      "headers": {
        "Authorization": "Bearer ${STALKR_API_KEY}"
      }
    }
  }
}

Tool results

Every successful tool call returns a short JSON text content block plus the same structured { data, meta } envelope as REST. Tool errors are marked isError and contain only a safe error code, message, and request ID.

The static resource stalkr://capabilities provides the live tool catalog and documentation link.

Safety

OAuth tokens are restricted to the exact MCP resource URL.
The MCP endpoint validates the public Host header to prevent DNS rebinding.
Every tool independently checks its scope and project role.
Trial and paid machine traffic use shared workspace rate buckets so multiple keys cannot multiply database load.
Destructive tools are annotated and still require explicit, validated input.