What it does
MCP Lens is a plugin for DeepSeek Harness that replaces a long list of MCP tool definitions with two model-facing tools. mcp_search finds relevant tools across servers and returns their exact input schemas, and mcp_call calls a chosen server and tool and returns the result, including structured data. According to the project's tests, the two definitions take 1,114 bytes of JSON whatever the catalog size, against 647,962 bytes for the direct client in its 1,000-tool fixture. Schemas enter the conversation only when a search returns them, connections open on demand and repeated searches reuse the catalog index. The latest version keeps identifiers returned with a text summary visible for later calls, and the project reports its tested data flows rising from 10/12 to 12/12. Servers, a cache path and allow and deny patterns are set in a Harness profile file. It supports MCP Tools over stdio and Streamable HTTP; OAuth, Resources, Prompts, Elicitation and task-based execution are not currently implemented. The project notes that search adds a step, so for a few tools used on nearly every request the official direct MCP client is simpler. It is MIT licensed and needs Node.js and a matching Harness version.
Capabilities
- Two model-facing tools, mcp_search and mcp_call
- Search across many MCP servers, with exact schemas loaded on demand
- Supports MCP Tools over stdio and Streamable HTTP
- Allow and deny patterns to limit which tools are exposed
- Structured results preserved for later calls
- Browser-based calculator for comparing schema size
When to use it
- Your MCP catalog has dozens to thousands of tools and crowds the model's context
- You use DeepSeek Harness and want to connect several MCP servers at once
- You want to measure how much standing schema size a tool catalog adds
