HasMCP's answer:
I was working on MCP Servers and I found it is hassle to create one using the tools/frameworks. But not only the creation process, the servers had be downloaded via npx or python at that time. As an alternative, I had to wrap an existing endpoint by writing tons of code. Another thing is the change on the official MCP specs was lightning fast and are not simple just to update your server, you have to write code again to catch it up with the latest spec.
After understanding the underlying protocol communication, I decided to remove this bottleneck from developers life and democratize to access MCP Server with just sharable URL instead of arbitrary local code execution.
Another big challenge was token optimization; I shouldn't have been just returning the what endpoint returns, that was including a lot of unnecessary sometimes confidential information, so an interceptor idea was born. HasMCP supports Jmespath and JS interceptors that you can modify the payload before returning to the client. Either you need to remove a PII column or just prune the data that you need, HasMCP is here to help.
Authentication is one of the other pain points in MCP world; HasMCP supports OAuth2 authentication with elicitation when the OAuth2 client credentials are defined in the provider. So, no more thinking about how I am going to authenticate the user. It is ready to use with just standard OAuth2 client input values. The access/refresh tokens are stored in encrypted format in the session storage. If your API does not support OAuth2, I got you covered too: Your users can define secrets (encrypted environment values) to send specific header values on tool calls.
Realtime access logs and metrics gives you full visibility on what is happening behind the scene. What is the request/response payload in JSONRPC 2.0 format, what MCP client/server side notifications have been sent. What methods have been called, all available in realtime only (not stored)!
When we expose MCPs to LLMs we give full control on the tools, another killer feature of HasMCP is toggling the tools what is being exposed to LLM in realtime. You can simply toggle the ones and authorize access for LLMs and this takes immediate affect, so you can kill a tool or add new one on the fly. This increases the security and give the user to have granular control what is allowed or not. While talking to one LLM you can expose a subset of read only tools whereas some superusers would need to use the write endpoints, this selection is just a toggle in HasMCP. Once you created an API provider, you can create unlimited servers from it, you can just focus on the secure design of your system.
HasMCP is API first, and majority of these functionality also available in opensource community edition. It is time to make your product 7/24 available to ChatGPT, Gemini, VSCode, Cursor, Claude, and all other LLMs that has MCP client with realtime governance. Today is the day to build an MCP server for your product!
JSFiddle
CodeSandbox
GitHub
replit
VS Code
Pastebin.com
JS Bin
FastMCP 3.0
Playground by Natoma
MCPForge.tech
Yavy
Mintlify
Hyperterse
