ContextForge MCP Gateway - the missing proxy for AI Agents and Tools
0 likes248 views
ContextForge MCP Gateway is a feature-rich gateway, proxy and MCP Registry that federates MCP and REST services - unifying discovery, auth, rate-limiting, observability, virtual servers, multi-transport protocols, and an optional Admin UI into one clean endpoint for your AI clients. It runs as a fully compliant MCP server, deployable via PyPI or Docker, and scales to multi-cluster environments on Kubernetes with Redis-backed federation and caching. It supports: Federation across multiple MCP and REST services Virtualization of legacy APIs as MCP-compliant tools and servers Transport over HTTP, JSON-RPC, WebSocket, SSE, stdio and streamable-HTTP An Admin UI for real-time management and configuration Built-in auth, observability, retries, and rate-limiting Scalable deployments via Docker or PyPI, Redis-backed caching, and multi-cluster federation
ContextForge MCP Gateway - the missing proxy for AI Agents and Tools
- 1. ContextForge MCP Gateway Open Source Gateway, Proxy and Registry for AI Agents and Tools Unifying discovery, authentication, authorization, role-based access control, rate-limiting, observability, virtual servers, and protocol translation Mihai Criveti – Distinguished Engineer, Agentic AI github.com/IBM/mcp-context-forge
- 2. Mihai Criveti | https://github.com/IBM/mcp-context-forge AI Agents use Tools to accomplish complex tasks AI Agent A goal-oriented autonomous system that plans and executes multi-step tasks, using LLMs to decide what to do next (planning) Uses an agentic framework or orchestrator (e.g. langgraph, langchain, autogen, crew.ai) + a LLM provider to execute tools and call external APIs. What can they do with tools? Retrieve external information (e.g. google, wikipedia, internal databases, RAG) as necessary, call other Agents or request user input Generate artifacts (images, diagrams, word documents, excel files), interact with external APIs and systems using various Tools, run unit tests, etc. AI Agent with planning, memory and tools Flow loop: Thought → Action → Observation → Repeat until task is complete Example coding agent that uses an internet browser to retrieve up-to-date library documentation, test code in a Code Interpreter and generate documents.
- 3. Mihai Criveti | https://github.com/IBM/mcp-context-forge Agents and Tools turn “thoughts” into actions Tools allow agents to take action in the world. Example tool: Name: google_search Description: Use this tool to search the internet, providing an input prompt. Input schema: {"query": "Latest News"} The LLM selects the right tool based on context, fills in the parameters, and the agentic framework invokes it — turning "thought" into real-world action.
- 4. Mihai Criveti | https://github.com/IBM/mcp-context-forge Challenges when developing Agent Tools (prior to MCP) Tools are small, purpose-built pieces of code designed to perform one task well — often tightly coupled to a particular platform or environment. Fragmented ecosystem: most tools are bound to specific frameworks or vendors (e.g. LangChain, Autogen), making cross-platform use difficult. Lack of standardization: each tool might use a different schema, naming convention, or prompting format — leading to inconsistency and complexity. Local-first execution: tools often run locally (in the same application as the Agent), not as reusable services — limiting scalability and accessibility. No shared discovery or registries: there’s no central way to find, share, or reuse tools across teams or platforms. Result: High friction, redundant code, limited interoperability. This is the gap MCP is designed to solve.
- 5. Mihai Criveti | https://github.com/IBM/mcp-context-forge Introducing Model Context Protocol (MCP) MCP Overview Open protocol introduced by Anthropic in November 2024 to standardize tool calling: “USB-C for AI tools” Enables a consistent interface to define how Agents and applications discover, invoke, and interact with tools and other context (prompts, resources) Widely adopted: 15,000 community servers developed since launch; wide adoption by major vendors. Still Evolving: the standard is rapidly evolving to improve security, granular access controls, transparent tool usage, and user interaction, but many tools only implement a partial or older spec. MCP Hosts: applications or AI tools that want to access data via MCP (Visual Studio Code + Copilot/Cline/Continue, Langchain, Crew.AI, Claude Desktop) MCP Clients: Protocol clients that maintain 1:1 connections with servers MCP Servers: Lightweight programs that each expose specific capabilities through Model Context Protocol
- 6. Mihai Criveti | https://github.com/IBM/mcp-context-forge MCP and the ecosystem are still maturing Current MCP challenges MCP gained rapid adoption — but the ecosystem evolved unevenly. Many tools only partially implement the spec, and integration challenges remain: Existing tools: REST endpoints must be rewritten to become MCP-compliant. Protocol and security inconsistency: some use JWT authentication, others OAuth2, many use nothing at all –while most servers are still developed to stdio / SSE transport (instead of the newer streamable HTTP). Integration complexity: tools are scattered across servers, each with its own config, retry logic, and monitoring gaps MCP Gateway emerges as a deployment pattern To address these limitations, we've introduced an open source project: ContextForge MCP Gateway. It acts as a smart proxy between clients and tool servers, solving integration pain points: • Acts as a secure, unified proxy between AI agents and tool servers, easily deployable in remote environments. • Convert between different transports, add authorization, security, observability, retry logic, and tool lifecycle controls • Wrap any REST API and exposes it as a typed, discoverable MCP-compatible tool https://github.com/IBM/mcp-context-forge
- 7. "ContextForge" MCP Gateway aggregates your MCP Servers: MCP Gateway Registry Virtual Servers Authorization Authentication Protocol Conversion -> any to any (stdio, SSE, Streamable HTTP, JSON- RPC, REST) Observability Rate Limiting HA / Routing Healthchecks API / UI / CLI Mihai Criveti | https://github.com/IBM/mcp-context-forge
- 8. MCP Gateway Architecture Mihai Criveti | https://github.com/IBM/mcp-context-forge https://github.com/IBM/mcp-context-forge 1. Register MCP Servers or REST APIs 2. Test tools with UI / APIs across multiple protocols 3. Create Virtual Servers with tools from multiple MCP Clients, exposing just the Tools/Resources and Prompts you need 4. Expose as REST, SSE, Streamable HTTP, STDIO, etc.