McProxy
A local, multi-tenant MCP server framework for macOS that consolidates multiple third-party services behind a single MCP endpoint, with a browser-based connection manager. It provides a secure, extensible foundation for adding custom connectors without shipping any built-in integrations.
README
McProxy
A local, multi-tenant Model Context Protocol server framework for
macOS, paired with a browser-based connection manager at 127.0.0.1:7337. It puts several
third-party services behind one MCP server, so a client like Claude Desktop or Claude Code sees one
process instead of one per integration.
What it deliberately does not include
McProxy ships zero connectors. src/connectors/descriptors/ is empty and the descriptor
registry is an empty array — there is nothing to configure, nothing to authorize, and no tool beyond
ping until you add one. This is the framework's central decision, not an oversight: shipping
someone else's integration would ship the shape of their credentials too. Every connector — its
fields, its auth flow, its probe — is added per install, either by a human following a guide or by an
AI agent running a setup interview.
Quick start
git clone https://github.com/DykstraSound/McProxy.git
cd McProxy
npm install
cp .env.example .env
npm run gui
npm run gui opens the connection manager in your browser. With no connectors configured yet, the
page has nothing to show — that's expected. The MCP server itself (npm run dev, or dist/server.js
after npm run build) also runs fine at this point; it answers ping and nothing else until a
connector exists.
Adding your first connection
See docs/ADDING-A-CONNECTOR.md to write a connector descriptor
yourself, or docs/AGENT-SETUP.md to have an AI agent walk you through it.
The descriptor model
Every connector is one file exporting one object: fields, auth kind, a live probe, an optional
identity check, and an optional reconnect flow. The GUI, the .env resolver, and the tool-layer
account scoping are all generic over this one shape — adding a connector never touches any of them.
export default defineConnector({
id: "example",
displayName: "Example",
kind: "native",
envPrefix: "EXAMPLE", // {PREFIX}_ACCOUNTS / {PREFIX}_{SLUG}_{FIELD}
multiInstance: true,
manageable: true,
auth: "oauth",
fields: [{ key: "API_KEY", label: "API key", secret: true, required: true }],
async probe(inst) { /* live reachability check — never rejects */ },
async identity(inst) { /* real remote account, for the identity-mismatch check */ },
authorize(inst) { return { script: "scripts/authorize-example.ts", args: [inst.slug] }; },
});
What it guarantees
- Third-party content is fenced before it reaches a model. Every tool result that carries data
fetched from a connected service is wrapped in a nonce-tagged envelope, minted after the payload
exists so nothing in it can forge a matching close marker (
src/shared/tools/untrusted.ts). - Secrets never leave the server through the API. The connection list masks every field marked
secretto a four-character prefix/suffix and the write-side form never echoes a stored value back (src/gui/api.ts,maskSecret/fieldViews). .envis written by exactly one module, which preserves comments and untouched lines byte for byte and takes a timestamped backup before every write (src/connectors/env-file.ts).- Every file under
~/.mcproxyis 0600 and every directory is 0700, enforced on every server and GUI start, not just at creation (src/shared/secure-fs.ts). - A probe never answers from a cache. Instance resolution is never memoized — a
.envedit is live on the next call, not the next restart (src/connectors/instances.ts) — and the cached status a connection displays is written only by an explicit Test or reconnect, never inferred (src/connectors/status.ts). - Absence is reported as absence, never as health. A connector nobody has tested renders
never-checked, not a default "ok" (src/connectors/status.ts); a config change is detected by re-resolving live state, never assumed from what was last observed.
Requirements
macOS, Node ≥ 22. Platform-agnostic support is an explicit non-goal.
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.