forge-mcp-server

What problem does it solve?

It prevents LLM agents from breaking or misusing MCP servers by enforcing the protocol and design constraints that make tool discovery, calling, and recovery reliable.

Core Features & Use Cases

• Transport correctness for MCP: Choose stdio for local-only integrated tools and HTTP/SSE streaming for remote, while ensuring stdio never writes to stdout.
• Capabilities that match reality: Declare only the MCP capabilities you actually implement to avoid confusing discovery and dead-end error paths.
• Agent-recoverable tool contracts: Use input schemas, structured error responses with codes, idempotent retries, permission scoping, and pagination/job patterns for large or long-running operations.
• Production-grade operational safety: Enforce auth for remote servers, log tool calls with hashed arguments, and avoid PII leakage while tracking error rates.
• Worked workflow support: Provides a minimal stdio server skeleton and a remote HTTP pattern, plus a verifier script to catch common hazards.

Quick Start

Use the forge-mcp-server skill to design your next MCP server and ensure its tools are discoverable, correctly schematized, and return structured errors the model can recover from.