We kept building the same thing. So we built the tool that builds it for you.
We're a small AI company in Freiburg, Germany. We build diagnostic systems, intelligent knowledge bases, and AI-powered hotlines for industrial enterprises.
Our clients are not Silicon Valley startups. They're manufacturing companies, logistics firms, pharmaceutical groups. Their tech stack isn't shiny — it's battle-tested. ERP platforms from the mid-2000s. SOAP services that nobody wants to touch. Databases that have outlived three CTOs.
And for the past year, every single one of them has been asking the same question:
"Can we just talk to our systems through AI?"
The Problem No One Talks About
The Model Context Protocol (MCP) is quickly becoming the standard for connecting AI agents to external systems. Claude, ChatGPT, Copilot, Cursor, Gemini — they all support it now.
But here's the dirty secret: every API needs its own custom MCP server.
Want to connect your ERP to Claude? Build a server. Your database? Another one. That old SOAP service from 2008 that handles compliance data? Yet another one.
For each server, you write the same boilerplate:
- Authentication handling
- Parameter mapping (path, query, body, headers)
- Error handling and retries
- Deployment and monitoring
- Documentation
We did this 15 times for different clients. Same patterns. Different APIs. Pure copy-paste engineering.
After the 15th integration, something had to change.
The Turning Point
We had a team meeting. Someone asked a simple question:
"We keep solving the same problem for every client. What if we build ONE tool that handles all of them?"
That conversation became a weekend project. The weekend project became a month-long effort. And the result became AnythingMCP — an open-source, self-hosted MCP gateway that turns any API into AI-ready tools.
No SDK. No code changes. No custom servers. Just point, configure, and connect.
How It Works
The concept is straightforward. AnythingMCP sits between your AI clients and your APIs:
Claude / ChatGPT / Copilot → AnythingMCP → Your APIs
You create a "connector" through a visual dashboard, point it at your API, and AnythingMCP generates MCP-compliant tools that any AI client can use.
What You Can Connect
REST APIs — The most common case. You can import directly from an OpenAPI/Swagger spec, a Postman collection, or even a raw cURL command. AnythingMCP parses the spec, creates tools with proper parameter mapping, and you're done.
SOAP/WSDL Services — This was the hardest one to build, and honestly the reason we started the project. Enterprise SOAP services are everywhere in German industry. Nobody wants to write a custom MCP server that handles XML namespaces, WS-Security headers, and certificate authentication. AnythingMCP auto-parses the WSDL and generates tools from it.
GraphQL — Point it at your endpoint. AnythingMCP runs introspection, discovers your queries and mutations, and creates corresponding tools.
Databases — Connect PostgreSQL, MySQL, MariaDB, MSSQL, Oracle, MongoDB, or SQLite directly. AnythingMCP does schema introspection at connection time and generates parameterized query tools. Read-only by default, because letting an AI write to your production database without guardrails is a terrible idea.
Other MCP Servers — The bridge connector lets you aggregate multiple MCP servers behind a single endpoint. One gateway for all your AI clients.
The Part Our Clients Actually Care About
Our clients don't care about MCP. They don't care about protocol translation or parameter mapping schemas.
They care about this: "I open Claude, I ask a question about my inventory, and I get a real answer from our actual system."
That's it. That's the product.
What used to take us 2–3 days per client integration now takes about 15 minutes:
- Create a connector in the dashboard
- Import the API spec (or paste a cURL command)
- Configure authentication
- Connect the AI client to the
/mcpendpoint
The middleware becomes invisible. Which is exactly what good infrastructure should be.
Why Enterprise Features Matter
We built AnythingMCP for enterprise environments from day one. Not because "enterprise" sounds impressive on a landing page, but because that's literally where we deploy it.
Authentication: OAuth2 with PKCE and Client Credentials, Bearer tokens, API keys, Basic auth, WS-Security for SOAP, and client certificates. Because every client's IT department has a different opinion on how auth should work.
Audit Logging: Every single tool invocation is logged — input parameters, output data, duration, status code. When your client's compliance team asks "what did the AI access and when?", you need an answer.
Role-Based Access Control: Not every user should be able to call every tool. You can whitelist specific tools per role. The intern doesn't get access to the financial reporting API.
Environment Variables: Per-connector variable interpolation with {{API_KEY}} syntax. The values are resolved server-side and never exposed to the AI. Your secrets don't leak into the LLM context.
Self-Hosted: This runs on your infrastructure. Or your client's infrastructure. The data never leaves the building. For German industrial companies, this is non-negotiable.
A Real Example
Let me walk you through a real (anonymized) setup we did recently.
A manufacturing client had three systems they wanted AI to access:
- An ERP system (REST API, OpenAPI spec available)
- A SOAP-based compliance service (WSDL endpoint, WS-Security)
- A PostgreSQL database with production metrics
Before AnythingMCP, this would have been three separate MCP server projects. Maybe two weeks of work including testing and deployment.
With AnythingMCP:
- ERP connector: Imported the OpenAPI spec. 47 tools auto-generated. Configured OAuth2 Client Credentials. Time: 10 minutes.
- SOAP connector: Pointed at the WSDL URL. 12 tools generated from the service operations. Added WS-Security credentials. Time: 15 minutes.
- Database connector: Connection string, schema introspection, read-only mode. 23 query tools auto-generated from the schema. Time: 5 minutes.
Total: 30 minutes. One MCP endpoint. 82 tools. The client opened Claude Desktop, connected to the endpoint, and started asking questions about their own systems.
Why We Open-Sourced It
We kept running into other teams with the exact same problem. At conferences, on calls, in online communities:
"We have 20 APIs. We need AI to talk to them. We really don't want to build 20 MCP servers."
We considered keeping it proprietary. But honestly, the MCP ecosystem needs more infrastructure tooling, not more walled gardens. The value for us isn't in the gateway itself — it's in the enterprise deployments, the customizations, and the support we provide around it.
AnythingMCP is source-available under the Business Source License 1.1. That means:
- Free for internal use, personal projects, development, testing, and academic use
- Not permitted to resell as a hosted SaaS without a separate commercial license
- Automatically converts to Apache 2.0 on March 4, 2030
Try It Yourself
One-click cloud deploy:
Railway has a template that provisions everything (app + PostgreSQL) in about 2 minutes. Click the deploy button on the GitHub repo and you're running.
Self-hosted with Docker:
git clone https://github.com/HelpCode-ai/anythingmcp.git
cd anythingmcp
./setup.sh
The setup script walks you through everything interactively: domain configuration, SSL (via Caddy with auto Let's Encrypt), authentication mode, SMTP, Redis. All secrets are auto-generated. The first user to register becomes admin.
Tech stack for the curious: NestJS 11, Next.js 16, React 19, PostgreSQL 17, Prisma 7, TypeScript. Streamable HTTP transport for MCP.
What's Next
We're actively developing AnythingMCP alongside our client work. On the roadmap:
- Webhooks and event-driven tools
- Tool versioning and changelogs
- Connector marketplace (share and import connector templates)
- Cloud-hosted option for teams that don't want to self-host
- More import formats (HAR, Insomnia, Bruno)
Links
🔗 GitHub: github.com/HelpCode-ai/anythingmcp
🌐 Website: anythingmcp.com
🎬 Demo video: anythingmcp.com/en/video-promo
We're helpcode.ai — a small team from Freiburg, Germany building AI infrastructure for enterprise. If AnythingMCP saves you time, a ⭐ on GitHub genuinely helps us keep building.
What system would you connect first? I'd love to hear about your use case in the comments.
Tags: MCP, Model Context Protocol, Open Source, AI, Enterprise, API Gateway, Claude, ChatGPT, Developer Tools, Self-Hosted
Top comments (0)