helpcodeai/anythingmcp repository overview

⁠AnythingMCP: the open-source MCP gateway that becomes your company brain

⁠

English⁠ · Deutsch⁠ · 简体中文⁠ · 日本語⁠

AnythingMCP is an open-source, self-hosted MCP gateway that connects Claude, ChatGPT, Copilot and Gemini to the software your company runs and to the documents on your file servers, without writing an MCP server.
On top of the gateway it keeps your company brain: a knowledge graph of how your systems connect, AI skills and company memory, shared by every AI client and limited by each user's role.
785 ready connectors with 12,000+ tools, among them SAP, Odoo, weclapp and Amazon Seller; 45 need no API key.

16,000+ workspaces on AnythingMCP Cloud · 1.5M+ AI tool calls a month · listed in the official Claude directory and the ChatGPT Plugin Directory

Try the Cloud free for 7 days⁠ · Add to Claude⁠ · Add to ChatGPT⁠ · Run it yourself⁠ · Docs⁠ · Connector guides⁠ · Discussions⁠

One question in Claude, answered from Etsy, SAP and a logistics API.

https://github.com/user-attachments/assets/cc8c9ef3-11cf-4eab-aa4d-98472dc554b3⁠

⁠What your AI gets

A gateway forwards calls. A company brain also knows where your documents are, how your systems connect and what your company has already decided. AnythingMCP gives every AI client both, through one MCP endpoint:

The self-hosted Community edition is free under AGPL-3.0 for any use, including commercial use, for up to 3 active users, with unlimited connectors and MCP servers. More users, single sign-on, SCIM, File Search beyond the 30-day trial and support come with Enterprise, on request: contact us⁠. Administrators can try single sign-on and SCIM for 30 days under Settings → License. Existing Starter and Team licence keys keep working.

⁠Run it yourself

Requires Docker 24+ and openssl; on macOS, start Docker Desktop first. No clone needed:

mkdir anythingmcp && cd anythingmcp
curl -fsSLo docker-compose.yml \
  https://raw.githubusercontent.com/HelpCode-ai/anythingmcp/main/docker-compose.quickstart.yml
printf 'JWT_SECRET=%s\nENCRYPTION_KEY=%s\n' "$(openssl rand -hex 32)" "$(openssl rand -hex 32)" > .env
docker compose up -d

Open http://localhost:3000⁠ and register: the first account becomes admin. Your MCP endpoint is http://localhost:4000/mcp, the API docs are at http://localhost:4000/api/docs.

On amd64 the image pulls in about 30 s and the API is healthy 24 s later. The image is amd64 only for now; the compose file pins the platform, so it also runs on Apple Silicon under emulation.

The quickstart binds to 127.0.0.1 on purpose, because nothing in front of it terminates TLS. For an instance other people or a cloud AI client can reach, clone the repo and run ./setup.sh: it asks for a domain, gets certificates through Caddy, generates the secrets and sets the MCP auth mode. See the Deployment Guide⁠.

⁠How it works

1. Add a connector. Install an adapter from the catalog (a ready JSON definition for SAP, Odoo, DHL and the rest), or point AnythingMCP at your own OpenAPI spec, WSDL, GraphQL endpoint, Postman collection or database. Once configured in your workspace it is a connector, and each of its operations is an MCP tool.
2. Decide what the model sees. Rename and describe the tools in the visual editor, drop the fields that must not leave your network, and choose which roles may call which tools.
3. Hand one URL to your AI client. An MCP server is the endpoint you add to Claude, ChatGPT, Copilot, Gemini or Cursor. It exposes the connectors you assign to it, and nothing else.

⁠Connect any API, SOAP service or database

Most companies have no MCP servers yet. They have a REST API, an ERP, a SOAP service from 2009 and a database. Each of them becomes a set of MCP tools:

Tools register at runtime, without a restart. Per-connector {{VAR}} values are interpolated on the server and never shown to the AI.

⁠Connector catalog

785 adapters, exposing 12,000+ tools. Every one has a setup guide on anythingmcp.com/guides⁠, in seven languages. The full catalog by category⁠ lists every Cloud connector with links to its guides for Claude, ChatGPT, Copilot, Cursor and Gemini, and the connector marketplace⁠ has a page for each one.

An adapter is a single JSON file. That is why the catalog is this size, and why adding one is a reasonable first contribution. Missing yours? Request it⁠ (we prioritise by 👍) or build it⁠. Your ERP isn't listed, or it's a custom build? Connect its REST API, its SOAP services or its SQL database directly, read-only.

⁠Security and governance

Everything runs on your infrastructure, so you decide what leaves it. OAuth2, roles, per-tool permissions, response mapping and the audit log are in every edition; SSO and SCIM need an Enterprise licence (a 30-day trial is built in).

• Response mapping. Each tool declares exactly which fields reach the model: drop a customer's IBAN or an employee's salary, or name the fields to keep. The editor shows the before/after on a real response; a shipped adapter goes from 12,172 B to 1,072 B (−91%), so you also stop paying for those fields in the context window. A broken mapping returns the raw response by default; set "fallbackToRaw": false on tools whose fields must never travel, and the call fails instead.
• Read-only where it matters. Every tool carries MCP annotations (readOnlyHint, destructiveHint), derived from the operation and overridable per tool. Role-based tool whitelisting lets you publish an MCP server that can only read, which is how most people should start with an ERP. Database query tools run a single SELECT and block writes and stacked statements.
• Every auth scheme you will meet. OAuth2 (PKCE and Client Credentials), Bearer, API key, Basic, HMAC request signing, LOGIN_TOKEN⁠ and OAuth 1.0a. Credentials are encrypted at rest with AES-256-GCM.
• Audit log. Every tool call is recorded with input, output, duration and status in your own database, including the full upstream response the model never saw.
• SSO⁠ and SCIM⁠ (Enterprise). Entra ID, Google, Okta, Auth0 or any OIDC provider. Roles sync from your directory groups on every sign-in; disable someone in the directory and their workspace access and MCP API keys go with it.

{
  "transform": {
    "mode": "select",
    "fallbackToRaw": false,
    "exclude": ["customer.iban", "customer.taxId"],
    "select": { "order": "$.id", "total": "$.amounts.gross", "status": "$.state" }
  }
}

Response mapping reference⁠

⁠Use it from Claude, ChatGPT, Copilot and Gemini

The same MCP server works in every client that speaks MCP, so you build a connector once:

• Claude. Add the server URL as a custom connector under Customize → Connectors; it then works in Claude.ai, Claude Desktop and Claude Code. OAuth 2.0 is supported out of the box. On AnythingMCP Cloud you can also add it in one click from the Claude Directory⁠. Claude setup⁠ · custom connectors for Claude⁠ · every connector's Claude guide⁠
• ChatGPT. On AnythingMCP Cloud, the AnythingMCP plugin in the ChatGPT Plugin Directory⁠ answers questions from your company databases, read-only. For every connector, add the server under Plugins in ChatGPT (Add → Add custom MCP server), or use it as the tool layer of an Apps SDK app. ChatGPT setup⁠ · ChatGPT apps and connectors⁠ · every connector's ChatGPT guide⁠
• Meta Muse. In Muse, open Settings → Connectors → Add custom connector, paste the server URL and sign in to AnythingMCP. Muse setup⁠ · every connector's Muse guide⁠
• Copilot, Gemini, Cursor and other MCP clients: client setup guides⁠, and a guide for every connector in Copilot⁠, Gemini⁠ and Cursor⁠.

⁠Company brain: knowledge graph, AI skills and memory

Forwarding calls leaves the hard part to the agent: knowing which tool to call next, and what your business means by "open order". AnythingMCP learns both and hands them back as context, not as extra tool calls.

• Knowledge Graph. A per-workspace map of entities (customers, orders, products) and how they relate across connectors, built from tool definitions and real calls, and editable by hand. It stores field names and relationships, never values. Each server exposes it through a kg_how_to_obtain tool, so the agent can ask how to get from a Shopware order to a DHL tracking number.
• AI skills. Recurring usage turned into small rules ("today's revenue includes order statuses 2, 3 and 4") and step-by-step procedures, written by hand or learned from real requests and fixed errors, versioned, and app