IBMContextForge is a gateway that puts Model Context Protocol (MCP) servers, Agent-to-Agent (A2A) agents, and REST or gRPC APIs behind a single endpoint for AI clients. It adds central authentication, rate limiting, discovery, and tracing across those backends, and it runs as a compliant MCP server in its own right.
It is organized as three gateways plus a plugin layer. The tools gateway federates MCP servers, wraps REST services as MCP tools with automatic JSON Schema extraction, and translates gRPC services to MCP through server reflection. The agent gateway routes A2A traffic to external agents, including OpenAI-compatible and Anthropic agents. The API gateway handles auth, retries, rate limits, and reverse proxying for REST services. More than 40 plugins add transports, protocols, and integrations.
Prompts, resources, and tools share unified registries: prompts are Jinja2 templates with versioning and rollback, resources are URI-addressed with MIME detection and caching, and tools carry input validation and concurrency controls. An Admin UI manages configuration and shows a live log viewer, and it supports airgapped deployments.
ContextForge deploys from PyPI or as a container, and scales to multi-cluster Kubernetes setups with Redis-backed federation and caching. Deployment guides cover AWS, Azure, Google Cloud Run, IBM Cloud Code Engine, Kubernetes with Helm, and OpenShift. Cross-gateway routing forwards the user's bearer token so remote gateways keep role-based access control, and it is blocked by default until trusted domains are allowlisted.
Features
- MCP federation: combines many MCP and REST services into one interface, with a selectable MCP protocol version
- REST and gRPC virtualization: wraps non-MCP services as virtual MCP servers, with gRPC methods discovered through reflection
- A2A agent routing: connects external agents over the A2A protocol, including OpenAI-compatible and Anthropic agents
- Transports: HTTP, JSON-RPC, WebSocket, SSE, and Streamable HTTP, with stdio for server-side use
- TOON compression: compresses tool payloads sent through the tools gateway
- Unified registries: versioned Jinja2 prompt templates, cached URI-based resources, and validated tools
- Auth and limits: Basic, JWT, or custom auth, user-scoped OAuth tokens, retries, and rate limiting
- OpenTelemetry tracing: distributed traces plus token, cost, and model metrics sent to Phoenix, Jaeger, Zipkin, Tempo, DataDog, or New Relic
- Admin UI: real-time management, configuration, and log search and export, including airgapped deployments
- Plugins: 40+ plugins for extra transports, protocols, and integrations
- Scaling: Redis-backed caching and multi-cluster federation on Kubernetes
