Sylex Search

What MastadoonPrime/sylex-search MCP server does

MastadoonPrime/sylex-search MCP server provides an agent-facing discovery index for products, services, and businesses. Agents can submit natural-language searches and receive ranked results with fit scores, then inspect individual listings for pricing, features, integrations, and reviews. The index began with software packages but is designed to represent broader categories, including businesses and non-software services.

The repository describes an index of more than 14,000 entries. Its initial data sources include npm and crates.io packages, Wikidata records, MCP servers, GitHub repositories, PyPI packages, and a smaller set of SaaS products. Results are returned as structured data rather than rendered web pages.

How it works

The hosted service is available through an SSE endpoint at https://search.sylex.ai/sse. An MCP-compatible client connects to that endpoint and makes the server's tools available to an agent. The documented local option runs the npm package through npx.

Search uses SQLite full-text search with custom ranking. The README states that the server does not make LLM calls or per-query API calls, so ranking is deterministic and query processing does not incur token costs. A server card is also available at https://search.sylex.ai/.well-known/mcp/server-card.json for clients that need connection and tool metadata without opening an MCP session.

Setup and configuration

MastadoonPrime/sylex-search MCP server can be connected through an MCP client's SSE configuration:

{
  "mcpServers": {
    "sylex-search": {
      "url": "https://search.sylex.ai/sse"
    }
  }
}

The README specifically lists Claude Desktop and Cursor among compatible clients. It also documents integrations with LangChain, the OpenAI Agents SDK, Microsoft AutoGen, and CrewAI through their MCP or SSE adapters.

For local stdio execution, the documented command is:

npx sylex-search

Self-hosting requires installing the Python requirements and setting AC_SUPABASE_URL, AC_SUPABASE_KEY, TRANSPORT, and PORT. The supplied example uses SSE transport on port 8080 and starts src/server.py.

Tools and capabilities

The search tools support these workflows:

• search.discover finds ranked matches for a natural-language query.
• search.details retrieves a listing by ID with available product data.
• search.compare compares between two and five products.
• search.categories lists categories and subcategories with counts.
• search.alternatives returns similar products with relevance scores.
• search.feedback accepts issue reports and improvement suggestions.

Listing-management tools support registering a new product or business, claiming an existing listing with URL proof, and updating fields such as descriptions, pricing, and categories. manage.register returns an owner token. manage.list_mcp records MCP connection configuration so other agents can discover and connect to a listed MCP server.

The repository also documents search-workflow and product-evaluation prompts for guided discovery and evaluation sequences.

Limitations and notes

The material describes the index as growing, so coverage may vary by category and source. Although the schema supports any product, service, or business, the seed data is weighted toward software and developer-related entries. Search quality depends on the records available in the index; the server is not described as a general web search or live price-verification system.

The hosted service is documented as requiring no authentication or configuration. Self-hosting, however, requires Supabase URL and key variables. The project is licensed under AGPL-3.0. The README identifies SSE as the transport and lists version 0.1.2.