Guides for connecting AI clients to AnythingMCP, the connector types you can build, and frequently asked questions. New here? Start with the README and Get started in 60 seconds.
Per-adapter setup guides (English, German, Italian and more) live at anythingmcp.com/guides.
| Client | Guide | Transport |
|---|---|---|
| Claude Desktop / Claude Code | Setup → | Streamable HTTP |
| ChatGPT | Setup → | Streamable HTTP |
| Google Gemini | Setup → | HTTP / SSE |
| GitHub Copilot | Setup → | Streamable HTTP |
| Cursor | Setup → | Streamable HTTP |
| Any MCP client | Setup → | Streamable HTTP |
| Connector | Use case | Docs |
|---|---|---|
| REST | HTTP APIs, OpenAPI/Swagger, Postman | Guide → |
| SOAP | WSDL web services, WCF, legacy enterprise APIs | Guide → |
| GraphQL | GraphQL endpoints with introspection | Guide → |
| Database | PostgreSQL, MySQL, MariaDB, MSSQL, Oracle, MongoDB, SQLite | Guide → |
| MCP Bridge | Aggregate multiple MCP servers into one | Guide → |
| LOGIN_TOKEN auth | APIs that POST credentials → return long-lived bearer | Guide → |
| Topic | Description |
|---|---|
| API reference | Full REST API for connectors, tools, auth, audit |
| Tool definition format | Parameters, endpoint mapping, response mapping |
| Knowledge Graph & AI skills | Per-workspace data-relationship map served over MCP, plus AI skills composed into server instructions |
| Deployment guide | Docker, production setup, reverse proxy, env vars |
| Authentication | OAuth2, JWT, API keys, MCP auth modes |
| License FAQ | Plain-language AGPL explanation |
How do I create a custom connector for Claude?
Run AnythingMCP (self-hosted or Cloud), import your API spec or pick a pre-built adapter, then add the gateway URL in Claude under Settings → Connectors. No code required — the Claude guide walks through it in ~5 minutes.
Can I build a ChatGPT app from my existing API?
Yes. As of December 2025, OpenAI calls these apps in ChatGPT (the term now covers both interactive apps and data connectors), and they're built on MCP via the Apps SDK. AnythingMCP generates the MCP backend from your existing API — add it as an app/connector in ChatGPT, or use it as the tool layer of an Apps SDK app. See Turn your API into a ChatGPT app.
Do I need to know how to code?
No. Importing an OpenAPI/Postman/WSDL spec, editing tools in the visual editor, and connecting Claude or ChatGPT are all point-and-click. Code only comes into play if you want to contribute a new adapter (a single JSON file) or self-host with custom infrastructure.
What is an MCP server?
An MCP server exposes tools to an AI agent over the Model Context Protocol — an open standard from Anthropic. Once connected, the AI can call those tools to read data, run queries, or perform actions on your behalf. AnythingMCP is a self-hosted MCP server that wraps your existing APIs so you don't have to write one from scratch.
How is AnythingMCP different from writing my own MCP server?
You don't write code. AnythingMCP imports your OpenAPI / Postman / WSDL spec (or you point it at a database) and generates the MCP tools automatically. You also get auth, audit logging, RBAC, and a visual editor on top — features that would take weeks to build per service.
Is AnythingMCP really open source?
Yes. AnythingMCP is licensed under the GNU AGPL v3, an OSI-approved open-source license — the same model as Twenty, Cal.com, Grafana and Plausible. You can read, fork, modify, self-host and even offer it as a service; the AGPL's network copyleft simply requires that modifications to the software stay open. The only exception is code under ee/ directories (cloud-operator features), which is commercially licensed and not needed for self-hosting. See the License FAQ.
Is AnythingMCP free?
Yes. AnythingMCP is free software under the AGPL v3 — free for internal company use, personal use, development, testing, evaluation and academic use. If you modify it and run it as a network service for others, the AGPL requires you to make your modified source available to those users. For commercial licensing without copyleft obligations: info@helpcode.ai.
Can I self-host?
Yes — it ships as a Docker image and runs on your own infrastructure. Run ./setup.sh or use the Railway and DigitalOcean one-click installs. There's also a managed Cloud version if you'd rather not run it yourself.
What about SOAP and WSDL?
Built-in. AnythingMCP automatically parses WSDL documents and generates one MCP tool per SOAP operation. Useful for legacy enterprise APIs (SAP, Oracle, .NET WCF, banking middleware) that no AI client speaks natively.
Is MCP dead now that agents use CLI tools?
No — but the question conflates two problems. CLI is the right call when the model already knows the tool from training (git, docker, kubectl), the agent is acting for the builder, and a CLI actually exists. MCP wins when you need per-user auth, scoped permissions, audit logs, multi-tenant isolation, or SaaS integrations without a CLI. The mature pattern in 2026 is hybrid: CLI for local/dev tools, MCP for SaaS / multi-tenant / compliance-bound integrations. Full decision matrix on anythingmcp.com/vs/cli.
Can the AI access my production database directly?
Yes, with safety. PostgreSQL, MySQL, MariaDB, MSSQL, Oracle, MongoDB and SQLite are supported. Each tool is whitelisted, every invocation is audit-logged, and you can scope a connector to read-only credentials. See the Database Connector Guide.
How is auth handled?
OAuth2 (PKCE + Client Credentials), Bearer Token, API Key, Basic Auth, query-parameter auth, WS-Security and TLS client certificates are all supported. Credentials are stored AES-256-GCM encrypted at rest. Per-user MCP API keys are issued on top so each AI client gets its own key with usage tracking.