HasMCP

#MCP (model context protocol) allows changing the available tools to the #MCPclient with Streamable HTTP in realtime. **How does it work?** 1. #MCPServer sends tool_list_changed event notification to the client using Streamable HTTP (server is responsible for blocking any calls to not advertised tools) 2. MCP Client makes request to tools/list method 3. #LLM (#Agent, #AI) is able to use available tools https://lnkd.in/gKt_kf82

In November 2024, it was beginning of the #ModelContextProtocol. Everything was new, the concept was new, spec was changing day by day. But in the middle of 2025, it became a standard. First #Google then #OpenAI accepted the MCP standard. Protocol made significant changes like killing SSE and introducing Streamable HTTP. There were a few approaches that frameworks were introducing the functionality by looking into schema definition and available documentation. It was a hard time. That was the time my journey started. It was around February/March, that I created my first MCP server using Python library. It was extremely challenging for me to debug as a developer due to lack of logging. In November 2025, MCP became part of Linux Foundation. With this approach, a clear message is given by #Anthropic, this is for all of us not for #Claude. Let's improve the protocol all together. As a #developer, despite the great documentation, evolution by the framework contributors, still it is a pain to have a properly functioning MCP server on top of APIs. It is not just #LLM create you can use it in one shot. Pain#1: As you might know protocol support #STDIO and #StreamableHTTP. Majority of the servers today still use STDIO despite that they don't need to interact with filesystem but just call API endpoints. Think of this, you will use someone's code in your machine to call API and on your own machine? It is a big security gap. This brings the problem of #deployment of the product MCP servers. Pain#2: #debugging; it is still hard with the current frameworks. I can tell realtime logs with telemetry are mandatory to ensure your MCP server is functional. Assume that as as developer you are sharing your MCP server as a library, what will happen when you need to debug on someones computer? Will you transfer data from your user's computer. Where is the privacy? Pain#3: #security: Think on a scenario, you would like to exclude a tool immediately that you don't want to expose to LLMs but the rest of the use-cases. How are you going to block? A gateway right? Consider that you would like to exclude a PII column not the endpoint before it reaches to LLMs. Pain#4: API changes, #versioning, spec changes. It is hard to maintain all. Pain#5: Token optimization. This is another challenge for the API owners who care about their users. Some endpoints return MBs of data where user needs a few attributes. It is bloating context and makes LLMs to hallucinate. And consumes a lot of tokens. HasMCP is a (#opensource with AGPL v3) GUI #MCPframework that maps your API to 7/24 online MCP Server with Streamable HTTP. It allows developers to alter request/response payloads, filter, prune PII columns or any attributes with interceptors without any change on your API. HasMCP has built-in auth, realtime debug logs and telemetry in addition to the usage analytics. Launching today (27th January) in #ProductHunt https://lnkd.in/gTAuKuYj #mcp #mcpserver #skills

https://lnkd.in/gU7tyFWH

Convert your API into MCP Server within seconds with built-in auth, realtime logs and token optimization. See how it works in 30 seconds demo. #mcpserver #mcp #modelcontextprotocol