ContextForge (IBM): MCP gateway tập trung — federate nhiều MCP servers, REST APIs, gRPC services qua một managed endpoint với auth, rate limiting, tracing.
Install
# PyPI
pip install mcp-contextforge-gateway
# Docker
docker run ghcr.io/ibm/mcp-context-forge:1.0.0-RC-3
# Kubernetes (Helm)
helm install contextforge ./charts/contextforge
3 Gateway Layers
Tools Gateway — MCP servers + REST-to-MCP translation
Agent Gateway — A2A protocol + OpenAI/Anthropic agent routing
API Gateway — rate limiting + auth + retries + reverse proxy
gRPC → MCP Auto-Translation
# Service reflection auto-converts gRPC → MCP tools
# Không cần viết wrapper thủ công
Unified Registries
Prompts — Jinja2 templating + versioning
Resources — URI-based + caching + SSE updates
Tools — native hoặc adapted, input validation
Observability
# OpenTelemetry tracing — any OTLP backend
# Phoenix, Jaeger, Zipkin compatible
# Distributed tracing qua federated gateways
# Automatic instrumentation: tools + resources + operations
Security
JWT authentication
RBAC (role-based access control)
Cross-gateway routing via domain allowlist
Strong JWT secret + admin credentials required
Admin UI
HTMX + Alpine.js — real-time management, config, log monitoring. Airgapped support.
Khi nào dùng
- Nhiều MCP servers cần manage tập trung
- Cần auth/rate-limit trước khi expose MCP tools
- Enterprise: multi-cluster federation + auto-scaling
- Muốn OTel tracing xuyên suốt MCP tool calls
Source
https://github.com/IBM/mcp-context-forge · Apache-2.0