IBM/mcp-context-forge

An AI Gateway, registry, and proxy that sits in front of any MCP, A2A, or REST/gRPC APIs, exposing a unified endpoint with centralized discovery, guardrails and management. Optimizes Agent & Tool calling, and supports plugins.

What it solves

ContextForge is a registry and proxy designed to unify fragmented AI tool and agent ecosystems. It solves the problem of managing multiple disparate Model Context Protocol (MCP) servers, REST APIs, and gRPC services by federating them into a single, governed endpoint for AI clients.

How it works

ContextForge acts as a gateway layer that translates and routes requests between AI clients and various backends. It provides several specialized gateways:

  • Tools Gateway: Handles MCP, REST, and gRPC-to-MCP translation, allowing non-MCP services to be used as MCP tools.
  • Agent Gateway: Manages Agent-to-Agent (A2A) integration and routing for agents from providers like OpenAI and Anthropic.
  • API Gateway: Provides enterprise-grade infrastructure like rate limiting, authentication, and retries for REST services.

It utilizes a plugin system with over 40 plugins for extensibility and supports multiple transports including HTTP, JSON-RPC, WebSocket, and SSE. For observability, it integrates with OpenTelemetry (OTLP) to track token usage, costs, and performance across distributed services.

Who it’s for

Developers and enterprise AI architects who need to coordinate multiple AI agents and tools across a complex infrastructure, particularly those using the Model Context Protocol (MCP) to standardize their tool-calling capabilities.

Highlights

  • Protocol Translation: Automatically converts gRPC and REST APIs into MCP-compliant tools.
  • Unified Registry: Centralized management of prompts (using Jinja2 templates), resources, and tools.
  • Enterprise Governance: Built-in OAuth tokens, rate limiting, and centralized authentication.
  • Observability: Vendor-agnostic tracing via OpenTelemetry with support for backends like Phoenix, Jaeger, and Zipkin.
  • Scalable Deployment: Deployable via PyPI, Docker, or Kubernetes with Redis-backed caching and federation.