Overview

This MCP Server Proxy aggregates multiple Model Context Protocol (MCP) resource servers behind a single HTTP endpoint. Instead of clients keeping track of many individual MCP endpoints, the proxy exposes a unified surface that forwards requests to configured resource backends — simplifying discovery, access control, logging and routing.

The proxy is useful when you want centralized management of resources (datasets, embeddings, context stores, etc.) across environments, teams or tenants. It can be deployed at the edge or inside a private network and configured to route or rewrite incoming MCP requests to one or more backends, making it easier for client applications to consume model context without changing endpoints.

Features

Aggregate multiple MCP resource servers behind one HTTP endpoint
Centralized routing and path-prefix rewriting
Simple YAML/JSON configuration for backend servers
Support for running as a standalone binary or inside Docker
Optional CORS and TLS configuration for secure access
Lightweight proxying — forwards requests and returns responses with minimal transformation
Basic health and discovery endpoints for monitoring available resources

Installation / Configuration

Prerequisites: Go toolchain not required to run releases. You can run a pre-built binary or use Docker.

Install and run (example: Linux/macOS binary)

# download a release binary (replace version)
curl -LO https://github.com/TBXark/mcp-proxy/releases/latest/download/mcp-proxy-linux-amd64
chmod +x mcp-proxy-linux-amd64
./mcp-proxy-linux-amd64 --config ./config.yaml

Docker run

docker run -p 8080:8080 \
  -v $(pwd)/config.yaml:/app/config.yaml:ro \
  ghcr.io/tbxark/mcp-proxy:latest \
  --config /app/config.yaml

Example configuration (YAML)

listen: "0.0.0.0:8080"
cors:
  enabled: true
  allow_origins: ["*"]
resources:
  - id: "vector-db"
    base_url: "http://10.0.1.5:9000"
    prefix: "/vector"
  - id: "file-index"
    base_url: "http://10.0.1.6:9000"
    prefix: "/files"
health:
  path: "/health"

Configuration fields (quick reference)

Key	Type	Description
listen	string	Address to bind (host:port)
resources[].id	string	Logical name for the resource server
resources[].base_url	string	Backend server URL to forward requests to
resources[].prefix	string	Incoming request prefix mapped to this backend
cors.enabled	bool	Enable CORS headers
health.path	string	Health check endpoint path

Available Tools / Resources

Typical endpoints exposed by the proxy (implementation may vary slightly):

GET /health — Returns proxy health status
GET /v1/resources — Lists configured resource servers and prefixes
Proxying behavior: any request to /mcp/{prefix}/… is forwarded to the backend mapped for that prefix, preserving method, headers and body

Example: if the config maps “vector” to /vector and backends vector-db, a request to GET /mcp/vector/items/123 will be forwarded to http://10.0.1.5:9000/items/123

Use the discovery endpoint to programmatically enumerate available resources:

curl http://localhost:8080/v1/resources

Use Cases

Centralized SDK endpoint for clients
- Problem: multiple microservices expose separate MCP resource APIs and clients must be configured with different URLs.
- Solution: deploy the proxy and configure each backend with a stable prefix. Clients only need to call the proxy URL; the proxy routes to the appropriate backend.
Staging vs Production routing
- Use different resource entries or weights to route traffic for testing or blue/green deployments. Change proxied base_url or prefix centrally without touching clients.
Access control and monitoring
- Insert authentication, request logging and CORS at a single location. Track usage across all MCP backends using the proxy access logs and metrics.
Local development and aggregation
- Run local MCP resource servers for embeddings, file indices, and context stores. Use the proxy to present a consistent API to local model runners and integration tests.

Concrete example: combining a vector DB and a file index

Configure two backends in config.yaml (vector-db -> /vector, file-index -> /files).
Start proxy on localhost:8080.
Vector lookup: curl http://localhost:8080/mcp/vector/search -d ‘{“q”:“query”}’
File metadata: curl http://localhost:8080/mcp/files/metadata/doc-123

Tips and Next Steps

Keep configuration in source control and use environment-specific config files for dev/staging/prod.
Use a reverse proxy (nginx/traefik) or load balancer in front for TLS termination if you need advanced TLS features.
Combine with observability (Prometheus, structured logs) to monitor backend health and latency across aggregated resources.

Repository and source: https://github.com/TBXark/mcp-proxy

MCP Server Proxy Aggregating MCP Resource Servers