An enterprise-grade AI Gateway by IBM that unifies MCP, A2A, and REST/gRPC APIs with centralized discovery, guardrails, and production-grade observability.
Core Capabilities#
ContextForge provides six categories of capabilities:
- Tools Gateway: Native MCP protocol support, REST/gRPC to MCP auto-conversion (reflection-based service discovery), TOON compression
- Agent Gateway: A2A (Agent-to-Agent) protocol, OpenAI compatible and Anthropic agent routing
- Model Gateway: LLM proxy with OpenAI API spec support, 8+ model providers
- API Gateway: Rate limiting, authentication, retry, reverse proxy
- Plugin Extensibility: 40+ plugins (transport, protocol, integrations)
- Observability: OpenTelemetry tracing with Phoenix, Jaeger, Zipkin and other OTLP backends
Transport & Protocol Support#
- Transports: HTTP, JSON-RPC, WebSocket, SSE (configurable keepalive), stdio, streamable-HTTP
- Protocols: MCP (native), A2A (Agent-to-Agent), REST, gRPC (reflection-based auto-conversion)
Deployment Options#
- PyPI:
pip install mcp-contextforge-gateway - Docker / Docker Compose: Full-stack deployment (PostgreSQL + Redis + Nginx)
- Kubernetes Helm: Multi-cloud (AWS, Azure, GCP, IBM Cloud, OpenShift) with HPA, RBAC, network policies
Quick Start#
# Single command with uv
JWT_SECRET_KEY=my-test-key \
MCPGATEWAY_UI_ENABLED=true \
MCPGATEWAY_ADMIN_API_ENABLED=true \
uvx --from mcp-contextforge-gateway mcpgateway --host 0.0.0.0 --port 4444
Admin UI: http://localhost:4444/admin API Docs: http://localhost:4444/docs (Swagger), http://localhost:4444/redoc
Security Features#
- Built-in authentication, retry, rate limiting
- User-level OAuth tokens
- X-Upstream-Authorization header for auth passthrough
REQUIRE_JTI=true(revocable tokens),REQUIRE_TOKEN_EXPIRATION=true(mandatory expiration)PUBLIC_REGISTRATION_ENABLED=false(public registration disabled)
Management & Extensibility#
- Admin UI: Real-time management, configuration, log monitoring, airgapped deployment support
- Plugin system: plugins/ and plugin_templates/ directories
- Redis federation and caching support
Tech Stack#
- Core Framework: FastAPI (Python)
- Data Layer: SQLAlchemy ORM (SQLite / PostgreSQL / MariaDB / MySQL)
- Language Composition: Python (83.9%), JavaScript (5.6%), HTML (5.1%), Makefile (2.0%), Rust (1.1%), Go (1.0%)
- Container Image: ghcr.io/ibm/mcp-context-forge
Core Author#
Mihai Criveti (Distinguished Engineer, Agentic AI, IBM)
Known Limitations#
- ARM64 production deployment not currently supported