Open Source · Self-Hosted · MCP Protocol

The self-hosted MCP hub
for AI assistants

Connect Claude, OpenCode, Cursor, and any MCP-compatible AI to all your tools through a single, secure endpoint - running entirely on your own infrastructure.

Get Started Read the Docs
1
Endpoint for all tools
0
Cloud dependencies
∞
Integrations possible
bash 
# Install the CLI
$ npm install -g mcp-depot

# Authenticate with your MCP Depot server (run once)
$ mcp-depot --login
  Server URL:  https://your-server.com
  API key:     mcd_xxxxxxxxxxxxxxxx
  Logged in successfully.

# Add to Claude Code
$ claude mcp add mcp-depot -- mcp-depot --mcp
  MCP server "mcp-depot" added.

Works with any MCP-compatible client

Claude Code
Claude Desktop
OpenCode
Cursor
Zed
Windsurf
Any MCP Client
Why MCP Depot

Built because we needed it ourselves

Cloud-hosted MCP aggregators require sending credentials to third-party services. We wanted something that runs on our own machines - and is useful enough to share.

Your credentials never leave your infra

API keys, tokens, and secrets stay in your Docker container. No cloud middleman, no data leaving your network. Full audit trail on your own logs.

One endpoint, every tool

Add HTTP APIs, connect external MCP servers (GitHub, Jira, Bitbucket), build custom integrations - all accessible to any AI client through a single MCP endpoint.

Teams collaborate in real time

Multi-user with JWT auth. Share context snapshots between sessions, post to communication channels, see which AI clients are connected - built for team workflows.

Features

Everything you need to run AI in production

From tool management to team collaboration - MCP Depot covers the full workflow.

HTTP Tool Builder

Turn any REST API into an MCP tool through the admin UI. Define endpoint, method, parameters, and auth - no code needed.

CORE

MCP Server Proxying

Connect external MCP servers (Bitbucket, GitHub, Jira, and more) via stdio or HTTP. All proxied through your unified endpoint.

CORE

Session Contexts

Named, persistent context snapshots. Store what you're working on, share it across AI sessions, pin important ones to never expire.

COLLABORATION

Session Channels

Append-only message channels between AI assistants and humans. Post updates, share findings, coordinate work across multiple sessions.

COLLABORATION

Connected Clients Panel

See exactly which AI clients are connected to your hub in real time - Claude Code, OpenCode, Cursor - with version and session info.

VISIBILITY

Multi-User Auth

JWT-based authentication with per-user API keys. Each team member gets their own access, with private and shared tool scopes.

SECURITY

Docker Compose Deploy

One command to start the full stack - server, database, and admin UI. No complex setup. Runs on any server or local machine.

DEVOPS

React Admin Dashboard

Manage tools, view connected clients, browse session channels and contexts - all from a clean, responsive web UI.

UI
How it works

From zero to connected in minutes

MCP Depot sits between your AI clients and your tools, acting as a proxy and hub for everything.

Step 01

Deploy on your server

Run docker compose up -d on any machine. MCP Depot starts with a React admin UI and REST API on port 3000.

Step 02

Add your tools and integrations

Use the admin dashboard to add HTTP API tools, connect external MCP servers, and configure credentials - all stored securely on your server.

Step 03

Point your AI client at it

Add MCP Depot to your AI client's config (Claude Desktop, Cursor, OpenCode) with your server URL and API key. Every tool you added is instantly available.

Architecture

How MCP Depot connects everything

One hub between your AI clients and all your tools. No per-client configuration, no credential sprawl.

Claude Desktop SSE / stdio Cursor HTTP transport OpenCode / Zed HTTP transport VS Code stdio / HTTP Any MCP Client any transport AI Clients MCP Depot Hub MCP Protocol Layer · stdio · SSE · HTTP/REST Tool Registry REST integrations Meta-tools API key auth Health monitoring Tool namespacing Skills & Personas Prompt templates Saved contexts Persona config Composite tools Skill sharing Session Manager Client tracking Tool monitoring Usage analytics Channels Context store PostgreSQL · persistent data store Integrations Jira Issue tracking GitHub / Bitbucket Code & PRs Confluence Docs & wikis Slack / Teams Messaging Custom HTTP APIs any endpoint
Quick Start

Running in under 5 minutes

MCP Depot ships as a Docker Compose stack. No external dependencies, no cloud accounts needed.

  • Clone the repo and run docker compose up -d
  • Create your admin account and generate an API key
  • Install the CLI and run mcp-depot --login
  • Add your first tool in the admin dashboard
  • Add the config snippet to your AI client
  • Start using your tools from any AI session
Server
CLI Login
Claude Desktop
Cursor
# Clone and start the server
$ git clone https://github.com/mcp-depot/mcp-depot
$ cd mcp-depot
$ docker compose up -d

# Admin UI available at http://localhost:3000
# Create your account and generate an API key
# Install the CLI on your machine
$ npm install -g mcp-depot

# Authenticate with your MCP Depot server
# (run once per machine - saves credentials locally)
$ mcp-depot --login

# Then start the MCP proxy for your AI client
$ mcp-depot --mcp
// First: mcp-depot --login  (one-time setup)
// Then add to claude_desktop_config.json:
// ~/Library/Application Support/Claude/
{
  "mcpServers": {
    "mcp-depot": {
      "command": "npx",
      "args": ["mcp-depot", "--mcp"],
      "env": {
        "MCP_DEPOT_URL":
          "http://localhost:3000",
        "MCP_DEPOT_API_KEY":
          "your-api-key"
      }
    }
  }
}
// First: mcp-depot --login  (one-time setup)
// Then add to .cursor/mcp.json in your project:
{
  "mcpServers": {
    "mcp-depot": {
      "command": "npx",
      "args":
        ["mcp-depot", "--mcp"],
      "env": {
        "MCP_DEPOT_URL":
          "http://localhost:3000",
        "MCP_DEPOT_API_KEY":
          "your-api-key"
      }
    }
  }
}
Roadmap

What's coming next

MCP Depot is actively developed. These are the features being built — shaped by real pain points teams face when deploying AI at scale.

Next major feature

Projects & Team Workspaces

A Project is a named bundle of integrations, tools, skills, and context - with its own membership. Assign developers to projects and they're instantly configured with everything they need.

1 Admin sets up Global project with company-wide tools (Jira, GitHub, Slack)
2 Team lead creates Project "Payments", adds Stripe + payment-debug skill
3 New developer joins - lead assigns them to the project
4 Developer: "set me up for Project Payments" → all tools and context loaded instantly

Projects also bring role-based access: admins onboard leads, leads onboard their own team members. Non-members can't see a project exists.

Tool Call Approval Workflows

Require human approval before sensitive tools execute - deploy, delete, send. AI makes the call, it waits in a queue, you approve or reject from the UI. Result flows back to the AI session.

PLANNED

Usage Analytics

Track which tools are called most, by which client and user, with what success rate and latency. Cost attribution per team when paired with the LLM proxy layer.

PLANNED

Webhook-Triggered Workflows

External events trigger pre-defined AI workflows automatically. Jira ticket created → run triage skill → post summary to Slack. No human needs to initiate the conversation.

PLANNED

PII & Data Governance

Scan tool inputs and outputs for sensitive data - emails, card numbers, API keys - before they are logged or forwarded. Redact, block, or flag. Required for regulated industries.

PLANNED

LLM Proxy Layer

Sit in front of model calls too - not just tool calls. Unified auth, rate limiting, system prompt injection, and full logging for all AI traffic through one control plane.

PLANNED

Integration Marketplace

Community-contributed integration bundles. One-click install of pre-configured tool sets for GitHub, Jira, Salesforce, Stripe, and more. MCP Depot becomes the registry for MCP tools.

PLANNED

Have a feature in mind? Shape the roadmap.

Join Discord GitHub Discussions Open an Issue