[tooling] · · 2 min read
mcp-proxy lets you bolt MCP onto any REST API without touching its code
A new open-source sidecar translates YAML-described REST endpoints into MCP tools, so legacy APIs can join the AI tool ecosystem in minutes.
By ByteBulletin Editors · Editorial Team
The Model Context Protocol is spreading fast, but most of the world's APIs are still plain REST. Rewriting them as MCP servers is a non-trivial project that most teams don't have time for. mcp-proxy, a new open-source tool from crhuber, aims to close that gap by acting as a translator: it reads a YAML config describing your existing REST endpoints and automatically exposes them as MCP tools, with zero changes to the upstream API.
The core idea is simple. You define one or more upstream base URLs in a config file, along with the tools you want to expose. mcp-proxy registers those tools with the MCP server, and when a client calls one, it makes the corresponding HTTP request to your real API and returns the result. In the project's example, a single config entry turns https://api.gold-api.com/price/XAU into a getPrice tool that any MCP client can call.
Config-driven tool generation
The config file is the heart of the project. Each tool has a name, description, and parameters. Parameters are mapped to upstream requests using a simple {parameter} syntax, and you can use them in paths, query strings, or request bodies. For instance, a customerId and currency property can be turned into a GET to /v1/invoices?customerId={customerId}¤cy={currency}.
Response handling is also configurable. Since MCP clients often prefer compact, structured text over raw JSON dumps, mcp-proxy includes a response.select option that uses JQ-style field selection. You can pull only the fields you need and present them in a markdown table or a bullet list, which is cheaper in tokens and easier for a model to parse.
Deployment and security
mcp-proxy is designed for a sidecar pattern, either in Kubernetes or in front of any API service that doesn't provide a native MCP server. It supports bearer token or header-based authentication for the upstream API, and it can protect the MCP server itself with an API key. One gotcha: the MCP SDK automatically enables DNS-rebinding protection, which can cause 403 errors if you're proxying from a loopback address with an external Host header. The project provides a flag, --disable-localhost-protection, for exactly that topology.
Building from source is straightforward, and the repo notes that the project can also be used with kelp. For teams weighed down by legacy REST services, mcp-proxy offers a pragmatic on-ramp to the MCP ecosystem — no rewriting business logic, just a YAML file and a sidecar.
SHARE
RELATED
[tooling] ·
AI-Proctored Remote Exam Fails So Badly That 58,000 Students Must Retake It In Person
An AI-supervised remote entrance exam at Mexico's largest university produced impossible score surges, forcing a historic in-person retest for tens of thousands of applicants.
[tooling] ·
local_bench: A zero-config TUI to benchmark the LLMs already on your machine
A new open-source tool turns your local model collection into a live terminal leaderboard measuring speed, memory, energy, and quality in one command.
[tooling] ·
OpenFOAM Agent: An AI Research Assistant for CFD Simulation Workflows
A new open-source agent framework aims to make OpenFOAM more accessible by automating setup, case preparation, and post-processing through natural language.