A TypeScript MCP server scaffold with one generated, classified tool per OpenAPI operation.
Grok MCP
Generate a Grok Bot connector from OpenAPI.
Sourced turns an OpenAPI spec into a Grok-ready remote MCP connector pack: server scaffold, allowlist, safety report, xAI SDK example, OpenAI-compatible Responses API example, cURL snippet, and hosted deployment checklist.
01Outputs
What Sourced generates
A Grok connector JSON file using server_url, server_label, server_description, and allowed_tools.
xAI native SDK and OpenAI-compatible SDK examples for the same remote MCP endpoint.
A cURL example against https://api.x.ai/v1/responses for teams that want to test outside an SDK.
A safety report that separates read, sensitive read, write, and destructive operations before Grok sees them.
02Details
The UX target
Upload the OpenAPI spec once.
Review the generated safe allowlist and deployment checklist in the same page.
Download the deployable pack, commit it to your own repo, and keep every external write owner-approved.
When a remote host needs a URL, paste only the deployed HTTPS /mcp endpoint. The connector files are generated for you.
03Details
Grok-specific safety rule
xAI remote MCP does not currently support OpenAI's require_approval or connector_id fields.
That means allowed_tools is the primary guardrail for Grok.
Sourced blocks the Grok config if the spec has OAuth/OpenID Connect setup gaps, an insecure authenticated base URL, or no safe read-only default tools.
Write and destructive operations stay out of the default Grok allowlist until an owner explicitly reviews them.
04Details
When to use it
Your customers are asking whether Grok Bot can call your API.
You already have OpenAPI and want a connector without hand-writing MCP handlers.
You want a generated package that can later be pushed into your own GitHub repo, not a random PR against xAI-owned repos.
You need a safety report before making a public remote MCP server reachable.