🤖 AI Summary
This post demonstrates that remote MCP (Model-Connected-Protocol) servers are just ordinary HTTP endpoints you can call with curl (or any HTTP client). The author provides a reusable cURL/JSON-RPC template: POST to the server URL with Content-Type: application/json and Accept: text/event-stream, application/json, and a body like {"jsonrpc":"2.0","id":"<SESSION_ID>","method":"tools/call","params":{"name":"<TOOL>","arguments":{...}}}. Responses come back as Server-Sent Events (SSE) whose data: lines contain JSON results (e.g., result.content = [{"type":"text","text":"..."}]). Example snippets show how to parse SSE with grep/sed/jq, how to stream/parse in JavaScript, and sample Ruby/GraphQL representations of the MCP content schema.
Why it matters: MCP tool invocations are a simple, standardized HTTP pattern that avoids proprietary SDK lock‑in and can be implemented in any language or pipeline. Important technical points: include authorization headers if required, any unique string can serve as the session id, handle SSE by filtering "data: " lines and ignoring "[DONE]", and extract JSON.result.content items. That makes it trivial to call a growing ecosystem of MCP tools from shells, servers, or client apps while preserving structured outputs (text, image, resource) for integration into ML workflows and orchestration layers.
Loading comments...
login to comment
loading comments...
no comments yet