ContextForge AI Gateway

ContextForge AI Gateway

An AI Gateway, registry, and proxy that sits in front of any MCP, A2A, or REST/gRPC APIs, exposing a unified control plane with centralized governance, discovery, and observability for all your AI needs. Optimizes Agent & Tool calling, and supports plugins.

ContextForge is a production-grade AI gateway, registry, and proxy that federates tools, agents, models, and APIs into one clean endpoint for your AI clients. It exposes a unified control plane with centralized governance, discovery, and observability across your entire AI infrastructure:

• Tools Gateway — MCP, REST, gRPC-to-MCP translation, and TOON compression
• Agent Gateway — A2A protocol, OpenAI-compatible and Anthropic agent routing
• Model Gateway — LLM proxy with OpenAI API spec supporting 8+ providers
• API Gateway — Rate limiting, auth, retries, and reverse proxy for REST services
• Plugin Extensibility — 40+ plugins for additional transports, protocols, and integrations
• Observability — OpenTelemetry tracing with Phoenix, Jaeger, Zipkin, and other OTLP backends

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.

Quick Links

!!! tip "Base URLs by runtime" - Direct installs (uvx, pip, or docker run): http://localhost:4444 - Docker Compose (nginx proxy): http://localhost:8080 - Dev server (make dev): http://localhost:8000

Overview & Goals

ContextForge is an AI gateway, registry, and proxy that sits in front of any Model Context Protocol (MCP) server, A2A server, or REST/gRPC API, exposing a unified control plane with centralized governance, discovery, and observability for all your AI needs. It optimizes agent and tool calling, and supports plugins. See the project roadmap for more details.

It currently supports:

• Federation across multiple MCP and REST services
• A2A (Agent-to-Agent) integration for external AI agents (OpenAI, Anthropic, custom)
• gRPC-to-MCP translation via automatic reflection-based service discovery
• Virtualization of legacy APIs as MCP-compliant tools and servers
• Transport over HTTP, JSON-RPC, WebSocket, SSE (with configurable keepalive), stdio and streamable-HTTP
• An Admin UI for real-time management, configuration, and log monitoring (with airgapped deployment support)
• Built-in auth, retries, and rate-limiting with user-scoped OAuth tokens and unconditional X-Upstream-Authorization header support
• OpenTelemetry observability with Phoenix, Jaeger, Zipkin, and other OTLP backends
• Scalable deployments via Docker or PyPI, Redis-backed caching, and multi-cluster federation

For a list of upcoming features, check out the ContextForge Roadmap

??? info "Gateway Layer with Protocol Flexibility"

* Sits in front of any MCP server or REST API
* Lets you choose your MCP protocol version (e.g., `2025-06-18`)
* Exposes a single, unified interface for diverse backends

??? info "Virtualization of REST/gRPC Services"

* Wraps non-MCP services as virtual MCP servers
* Registers tools, prompts, and resources with minimal configuration
* **gRPC-to-MCP translation** via server reflection protocol
* Automatic service discovery and method introspection

??? info "REST-to-MCP Tool Adapter"

* Adapts REST APIs into tools with:

    * Automatic JSON Schema extraction
    * Support for headers, tokens, and custom auth
    * Retry, timeout, and rate-limit policies

??? info "Unified Registries"

* **Prompts**: Jinja2 templates, multimodal support, rollback/versioning
* **Resources**: URI-based access, MIME detection, caching, SSE updates
* **Tools**: Native or adapted, with input validation and concurrency controls

??? info "Admin UI, Observability & Dev Experience"

* Admin UI built with HTMX + Alpine.js
* Real-time log viewer with filtering, search, and export capabilities
* Auth: Basic, JWT, or custom schemes
* Structured logs, health endpoints, metrics
* 400+ tests, Makefile targets, live reload, pre-commit hooks

??? info "OpenTelemetry Observability"

* **Vendor-agnostic tracing** with OpenTelemetry (OTLP) protocol support
* **Multiple backend support**: Phoenix (LLM-focused), Jaeger, Zipkin, Tempo, DataDog, New Relic
* **Distributed tracing** across federated gateways and services
* **Automatic instrumentation** of tools, prompts, resources, and gateway operations
* **LLM-specific metrics**: Token usage, costs, model performance
* **Zero-overhead when disabled** with graceful degradation

See **[Observability Documentation](manage/observability.md)** for setup guides with Phoenix, Jaeger, and other backends.

Quick Start - PyPI

ContextForge is published on PyPI as mcp-contextforge-gateway.

TLDR (single command using uv):

# Quick start with environment variables
JWT_SECRET_KEY=my-test-key \
MCPGATEWAY_UI_ENABLED=true \
MCPGATEWAY_ADMIN_API_ENABLED=true \
PLATFORM_ADMIN_EMAIL=admin@example.com \
PLATFORM_ADMIN_PASSWORD=changeme \
PLATFORM_ADMIN_FULL_NAME="Platform Administrator" \
uvx --from mcp-contextforge-gateway mcpgateway --host 0.0.0.0 --port 4444

# Or better: use the provided .env.example
cp .env.example .env
# Edit .env to customize your settings
uvx --from mcp-contextforge-gateway mcpgateway --host 0.0.0.0 --port 4444

??? note "Prerequisites"

* **Python ≥ 3.10** (3.11 recommended)
* **curl + jq** - only for the last smoke-test step

1 - Install & run (copy-paste friendly)

# 1️⃣  Isolated env + install from pypi
mkdir mcpgateway && cd mcpgateway
python3 -m venv .venv && source .venv/bin/activate
pip install --upgrade pip
pip install mcp-contextforge-gateway

# 2️⃣  Copy and customize the configuration
# Download the example environment file
curl -O https://raw.githubusercontent.com/IBM/mcp-context-forge/main/.env.example
cp .env.example .env
# Edit .env to customize your settings (especially passwords!)

# Or set environment variables directly:
export MCPGATEWAY_UI_ENABLED=true
export MCPGATEWAY_ADMIN_API_ENABLED=true
export PLATFORM_ADMIN_EMAIL=admin@example.com
export PLATFORM_ADMIN_PASSWORD=changeme
export PLATFORM_ADMIN_FULL_NAME="Platform Administrator"

JWT_SECRET_KEY=my-test-key \
  mcpgateway --host 0.0.0.0 --port 4444 &

# 3️⃣  Generate a bearer token & smoke-test the API
export MCPGATEWAY_BEARER_TOKEN=$(python3 -m mcpgateway.utils.create_jwt_token \
    --username admin@example.com --exp 10080 --secret my-test-key)

curl -s -H "Authorization: Bearer $MCPGATEWAY_BEARER_TOKEN" \
     http://127.0.0.1:4444/version | jq

??? example "Windows (PowerShell) quick-start"

```powershell
# 1️⃣  Isolated env + install from PyPI
mkdir mcpgateway ; cd mcpgateway
python3 -m venv .venv ; .\.venv\Scripts\Activate.ps1
pip install --upgrade pip
pip install mcp-contextforge-gateway

# 2️⃣  Copy and customize the configuration
Invoke-WebRequest -Uri "https://raw.githubusercontent.com/IBM/mcp-context-forge/main/.env.example" -OutFile ".env.example"
Copy-Item .env.example .env
# Edit .env to customize your settings

# Or set environment variables (session-only)
$Env:MCPGATEWAY_UI_ENABLED        = "true"
$Env:MCPGATEWAY_ADMIN_API_ENABLED = "true"
$Env:JWT_SECRET_KEY               = "my-test-key"
$Env:PLATFORM_ADMIN_EMAIL         = "admin@example.com"
$Env:PLATFORM_ADMIN_PASSWORD      = "changeme"
$Env:PLATFORM_ADMIN_FULL_NAME     = "Platform Administrator"

# 3️⃣  Launch the gateway
mcpgateway.exe --host 0.0.0.0 --port 4444

# 4️⃣  Bearer token and smoke-test
$Env:MCPGATEWAY_BEARER_TOKEN = python3 -m mcpgateway.utils.create_jwt_token `
    --username admin@example.com --exp 10080 --secret my-test-key

curl -s -H "Authorization: Bearer $Env:MCPGATEWAY_BEARER_TOKEN" `
     http://127.0.0.1:4444/version | jq
```

??? example "End-to-end demo (register a local MCP server)"

```bash
# 1️⃣  Spin up the sample GO MCP time server using mcpgateway.translate & docker
python3 -m mcpgateway.translate \
     --stdio "docker run --rm -i ghcr.io/ibm/fast-time-server:latest -transport=stdio" \
     --expose-sse \
     --port 8003

# Or using the official mcp-server-git using uvx:
pip install uv # to install uvx, if not already installed
python3 -m mcpgateway.translate --stdio "uvx mcp-server-git" --expose-sse --port 9000

# NEW: Expose via multiple protocols simultaneously!
python3 -m mcpgateway.translate \
     --stdio "uvx mcp-server-git" \
     --expose-sse \
     --expose-streamable-http \
     --port 9000
# Now accessible via both /sse (SSE) and /mcp (streamable HTTP) endpoints

# 2️⃣  Register it with the gateway
curl -s -X POST -H "Authorization: Bearer $MCPGATEWAY_BEARER_TOKEN" \
     -H "Content-Type: application/json" \
     -d '{"name":"fast_time","url":"http://localhost:8003/sse"}' \
     http://localhost:4444/gateways

# 3️⃣  Verify tool catalog
curl -s -H "Authorization: Bearer $MCPGATEWAY_BEARER_TOKEN" http://localhost:4444/tools | jq

# 4️⃣  Create a virtual server bundling those tools
curl -s -X POST -H "Authorization: Bearer $MCPGATEWAY_BEARER_TOKEN" \
     -H "Content-Type: application/json" \
     -d '{"server":{"name":"time_server","description":"Fast time tools","associated_tools":["<ID_OF_TOOLS>"]}}' \
     http://localhost:4444/servers | jq

# 5️⃣  List servers
curl -s -H "Authorization: Bearer $MCPGATEWAY_BEARER_TOKEN" http://localhost:4444/servers | jq

# 6️⃣  Client HTTP endpoint - use MCP Inspector
npx -y @modelcontextprotocol/inspector
# Transport Type: Streamable HTTP, URL: http://localhost:4444/servers/UUID_OF_SERVER_1/mcp
```

??? example "Using the stdio wrapper (mcpgateway-wrapper)"

```bash
export MCP_AUTH="Bearer ${MCPGATEWAY_BEARER_TOKEN}"
export MCP_SERVER_URL=http://localhost:4444/servers/UUID_OF_SERVER_1/mcp
python3 -m mcpgateway.wrapper  # Ctrl-C to exit
```

When using a MCP Client such as Claude with stdio:

```json
{
  "mcpServers": {
    "mcpgateway-wrapper": {
      "command": "python",
      "args": ["-m", "mcpgateway.wrapper"],
      "env": {
        "MCP_AUTH": "Bearer your-token-here",
        "MCP_SERVER_URL": "http://localhost:4444/servers/UUID_OF_SERVER_1",
        "MCP_TOOL_CALL_TIMEOUT": "120"
      }
    }
  }
}
```

Quick Start - Containers

Use the official OCI image from GHCR with Docker or Podman.

!!! note "ARM64 Support" Currently, arm64 is not supported in production. On macOS with Apple Silicon (M1, M2, etc), use Rosetta or install via PyPI instead.

Docker Compose (Recommended)

Get a full stack running with MariaDB and Redis:

# Clone and start the stack
git clone https://github.com/IBM/mcp-context-forge.git
cd mcp-context-forge

# Start with MariaDB (recommended for production)
docker compose up -d

# Check status
docker compose ps

# View logs
docker compose logs -f gateway

# Access Admin UI: http://localhost:4444/admin
# Generate API token
docker compose exec gateway python3 -m mcpgateway.utils.create_jwt_token \
  --username admin@example.com --exp 10080 --secret my-test-key

What you get:

• MariaDB 10.6 - Production-ready database with 36+ tables
• ContextForge - Full-featured gateway with Admin UI
• Redis - High-performance caching and session storage
• Admin Tools - pgAdmin, Redis Insight for database management
• Nginx Proxy - Caching reverse proxy (optional)

Helm (Kubernetes)

Deploy to Kubernetes with enterprise-grade features:

# Clone and use local chart
git clone https://github.com/IBM/mcp-context-forge.git
cd mcp-context-forge/charts/mcp-stack

# Install with MariaDB
helm install mcp-gateway . \
  --set mcpContextForge.secret.PLATFORM_ADMIN_EMAIL=admin@yourcompany.com \
  --set mcpContextForge.secret.PLATFORM_ADMIN_PASSWORD=changeme \
  --set mcpContextForge.secret.JWT_SECRET_KEY=your-secret-key \
  --set postgres.enabled=false \
  --set mariadb.enabled=true

# Check deployment status
kubectl get pods -l app.kubernetes.io/name=mcp-context-forge

# Port forward to access Admin UI
kubectl port-forward svc/mcp-gateway-mcp-context-forge 4444:80

Docker (Single Container)

docker run -d --name mcpgateway \
  -p 4444:4444 \
  -e MCPGATEWAY_UI_ENABLED=true \
  -e MCPGATEWAY_ADMIN_API_ENABLED=true \
  -e HOST=0.0.0.0 \
  -e JWT_SECRET_KEY=my-test-key \
  -e AUTH_REQUIRED=true \
  -e PLATFORM_ADMIN_EMAIL=admin@example.com \
  -e PLATFORM_ADMIN_PASSWORD=changeme \
  -e PLATFORM_ADMIN_FULL_NAME="Platform Administrator" \
  -e DATABASE_URL=sqlite:///./mcp.db \
  -e SECURE_COOKIES=false \
  ghcr.io/ibm/mcp-context-forge:1.0.0-RC-1

# Tail logs and generate API key
docker logs -f mcpgateway
docker run --rm -it ghcr.io/ibm/mcp-context-forge:1.0.0-RC-1 \
  python3 -m mcpgateway.utils.create_jwt_token --username admin@example.com --exp 10080 --secret my-test-key

Browse to http://localhost:4444/admin and login with PLATFORM_ADMIN_EMAIL / PLATFORM_ADMIN_PASSWORD.

??? example "Advanced: Persistent storage, host networking, airgapped"

**Persist SQLite database:**

```bash
mkdir -p $(pwd)/data && touch $(pwd)/data/mcp.db && chmod 777 $(pwd)/data
docker run -d --name mcpgateway --restart unless-stopped \
  -p 4444:4444 -v $(pwd)/data:/data \
  -e DATABASE_URL=sqlite:////data/mcp.db \
  -e MCPGATEWAY_UI_ENABLED=true -e MCPGATEWAY_ADMIN_API_ENABLED=true \
  -e HOST=0.0.0.0 -e JWT_SECRET_KEY=my-test-key \
  -e PLATFORM_ADMIN_EMAIL=admin@example.com -e PLATFORM_ADMIN_PASSWORD=changeme \
  ghcr.io/ibm/mcp-context-forge:1.0.0-RC-1
```

**Host networking** (access local MCP servers):

```bash
docker run -d --name mcpgateway --network=host \
  -v $(pwd)/data:/data -e DATABASE_URL=sqlite:////data/mcp.db \
  -e MCPGATEWAY_UI_ENABLED=true -e HOST=0.0.0.0 -e PORT=4444 \
  ghcr.io/ibm/mcp-context-forge:1.0.0-RC-1
```

**Airgapped deployment** (no internet):

```bash
docker build -f Containerfile.lite -t mcpgateway:airgapped .
docker run -d --name mcpgateway -p 4444:4444 \
  -e MCPGATEWAY_UI_AIRGAPPED=true -e MCPGATEWAY_UI_ENABLED=true \
  -e HOST=0.0.0.0 -e JWT_SECRET_KEY=my-test-key \
  mcpgateway:airgapped
```

Podman (rootless-friendly)

podman run -d --name mcpgateway \
  -p 4444:4444 -e HOST=0.0.0.0 -e DATABASE_URL=sqlite:///./mcp.db \
  ghcr.io/ibm/mcp-context-forge:1.0.0-RC-1

VS Code Dev Container

Clone the repo and open in VS Code—it will detect .devcontainer and prompt to "Reopen in Container". The container includes Python 3.11, Docker CLI, and all project dependencies.

For detailed setup, workflows, and GitHub Codespaces instructions, see Developer Onboarding.

Installation

make venv install          # create .venv + install deps
make serve                 # gunicorn on :4444

??? example "Alternative: UV or pip"

```bash
# UV (faster)
uv venv && source .venv/bin/activate
uv pip install -e '.[dev]'

# pip
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
```

??? example "PostgreSQL adapter setup"

Install the `psycopg` driver for PostgreSQL:

```bash
# Install system dependencies first
# Debian/Ubuntu: sudo apt-get install libpq-dev
# macOS: brew install libpq

uv pip install 'psycopg[binary]'   # dev (pre-built wheels)
# or: uv pip install 'psycopg[c]'  # production (requires compiler)
```

Connection URL format:

```bash
DATABASE_URL=postgresql+psycopg://user:password@localhost:5432/mcp
```

Upgrading

For upgrade instructions, migration guides, and rollback procedures, see:

• Upgrade Guide — General upgrade procedures
• CHANGELOG.md — Version history and breaking changes
• MIGRATION-0.7.0.md — Multi-tenancy migration (v0.6.x → v0.7.x)

Configuration

!!! warning "Startup Validation" If any required .env variable is missing or invalid, the gateway will fail fast at startup with a validation error via Pydantic.

Copy the provided .env.example to .env and update the security-sensitive values below.

Required: Change Before Use

These variables have insecure defaults and must be changed before production deployment:

Security Defaults (Secure by Default)

These settings are enabled by default for security—only disable for backward compatibility:

Project Defaults (Dev Setup)

These values differ from code defaults to provide a working local/dev setup:

Full Configuration Reference

For the complete list of 300+ environment variables organized by category (authentication, caching, SSO, observability, etc.), see the Configuration Reference.

Running

Quick Reference

Development Server (Uvicorn)

make dev                 # Uvicorn on :8000 with auto-reload and SQLite
# or
./run.sh --reload --log debug --workers 2

run.sh is a wrapper around uvicorn that loads .env, supports reload, and passes arguments to the server.

Key flags:

Production Server (Gunicorn)

make serve               # Gunicorn on :4444 with multiple workers
make serve-ssl           # Gunicorn behind HTTPS on :4444 (uses ./certs)

Docker Compose (Full Stack)

make compose-up          # Start full stack: PostgreSQL, Redis, 3 gateway replicas, Nginx on :8080
make compose-logs        # Tail logs from all services
make compose-down        # Stop the stack

Manual (Uvicorn)

uvicorn mcpgateway.main:app --host 0.0.0.0 --port 4444 --workers 4

Cloud Deployment

ContextForge can be deployed to any major cloud platform:

For comprehensive deployment guides, see Deployment Documentation.

API Reference

Interactive API documentation is available when the server is running (adjust the base URL for Compose or dev):

• Swagger UI (http://localhost:4444/docs or http://localhost:8080/docs with Compose) — JWT-protected by default; enable DOCS_ALLOW_BASIC_AUTH=true or log in to the Admin UI to get a session cookie.
• ReDoc (http://localhost:4444/redoc or http://localhost:8080/redoc with Compose) — Same auth requirements as Swagger UI.

Quick Authentication:

# Generate a JWT token
export TOKEN=$(python3 -m mcpgateway.utils.create_jwt_token \
  --username admin@example.com --exp 10080 --secret my-test-key)

# Test API access
curl -H "Authorization: Bearer $TOKEN" http://localhost:4444/health

For comprehensive curl examples covering all endpoints, see the API Usage Guide.

Testing

make test            # Run unit tests
make lint            # Run all linters
make doctest         # Run doctests
make coverage        # Generate coverage report

See Doctest Coverage Guide for documentation testing details.

Project Structure

mcpgateway/          # Core FastAPI application
├── main.py          # Entry point
├── config.py        # Pydantic Settings configuration
├── db.py            # SQLAlchemy ORM models
├── schemas.py       # Pydantic validation schemas
├── services/        # Business logic layer (50+ services)
├── routers/         # HTTP endpoint definitions
├── middleware/      # Cross-cutting concerns
└── transports/      # SSE, WebSocket, stdio, streamable HTTP

tests/               # Test suite (400+ tests)
docs/docs/           # Full documentation (MkDocs)
charts/              # Kubernetes/Helm charts
plugins/             # Plugin framework and implementations

For complete structure, see CONTRIBUTING.md or run tree -L 2.

Development

make dev             # Dev server with auto-reload (:8000)
make test            # Run test suite
make lint            # Run all linters
make coverage        # Generate coverage report

Run make to see all 75+ available targets.

For development workflows, see:

• Developer Workstation Setup
• Building & Packaging

Troubleshooting

Common issues and solutions:

For detailed troubleshooting guides, see Troubleshooting Documentation.

Contributing

1. Fork the repo, create a feature branch.
2. Run make lint and fix any issues.
3. Keep make test green.
4. Open a PR with signed commits (git commit -s).

See CONTRIBUTING.md for full guidelines and Issue Guide #2502 for how to file bugs, request features, and find issues to work on.

Changelog

A complete changelog can be found here: CHANGELOG.md

License

Licensed under the Apache License 2.0 - see LICENSE