MCP server search engine for AI Agents

[sisambam]

Pre-submission Checklist

• I have verified this would not be more appropriate as a feature request in a specific repository
• I have searched existing discussions to avoid duplicates

Your Idea

Problem

AI agents today can connect to MCP servers which expose tools, prompts, and resources, allowing them to interact with external APIs, databases, and document stores. However, two key gaps remain:

1. Content Search: There is no unified MCP-native way to perform search across heterogeneous resources (documents, APIs, the web) with a consistent query/response interface.
2. Server Discovery: Agents cannot easily discover new MCP servers and their capabilities. Instead, they rely on pre-configured connections, limiting their reach and adaptability.

Without a standard for either of these, the “Agentic Web” remains fragmented and static.

Who benefits

1. AI Agents (primary end-users): Gain the ability to search both content (papers, docs, APIs) and infrastructure (new MCP servers), mirroring how humans use search to expand knowledge.
2. Human Users (ultimate beneficiaries): Experience more relevant, accurate, and up-to-date agent responses.
3. Agent Developers: Avoid reinventing retrieval and discovery logic; can plug into standardized search capabilities.
4. Ecosystem Contributors: Can extend the model with new backends (Elasticsearch, Postgres FTS, vector DBs, hybrid RAG systems) or register new MCP servers, while maintaining interoperability.

Proposed solution

1. MCP Search Server (Content Search)
   Introduce an MCP server that exposes a search tool. This allows agents to query across text, structured data, or APIs, and receive ranked results with metadata.
   Example request
   { "name": "search", "arguments": { "query": "clinical trial matching with AI", "filters": { "source": "pubmed", "date_range": { "from": "2023-01-01T00:00:00Z", "to": "2025-09-27T00:00:00Z" } }, "top_k": 3 } }
   Example response
   { "results": [ { "id": "doc_001", "title": "Artificial Intelligence for Clinical Trial Matching", "snippet": "This study evaluates how AI models improve patient-to-trial matching...", "url": "https://pubmed.ncbi.nlm.nih.gov/12345678", "score": 0.93, "source": "pubmed", "metadata": { "author": "J. Smith et al.", "published_at": "2024-05-12T00:00:00Z" } } ], "next_page_token": "page_2_token" }

2. MCP Server Discovery (Registry Search)

Extend the same concept to server-level search: a directory where agents can find new MCP servers and their advertised capabilities.
Example Use Case
Agent asks: “Find me a weather data server.”
Search engine returns:
{ "servers": [ { "id": "weather_001", "name": "OpenWeather MCP Server", "capabilities": ["getForecast", "getAlerts"], "url": "https://mcp.example.com/weather" } ] }

Scope

• Protocol Specification
• SDK Features
• Documentation
• Developer Experience
• Other

[aarora79]

Maybe some ideas from here would be relevant https://github.com/agentic-community/mcp-gateway-registry/tree/main, specifically the intelligent_tool_finder https://github.com/agentic-community/mcp-gateway-registry/blob/main/docs/dynamic-tool-discovery.md.

[sisambam]

This is great, the gap AFAICT between my proposal and the one here is

1. the tool finder works on a per registry level - am talking about a search tool across registries.
2. the tool finder finds tools only - need support for prompts and resources.
   But I guess we can expand the existing tool to fill these gaps.