agentic-search
A Model Context Protocol (MCP) server that provides agentic search capabilities with support for vector search using Qdrant, full-text search using TiDB, or both combined.
claude mcp add --transport stdio cardea-mcp-agentic-search npx -y cardea-mcp-agentic-search \ --env TiDB_DSN="Data Source Name for TiDB" \ --env TiDB_DATABASE="Database name to use" \ --env TiDB_PASSWORD="Password for TiDB" \ --env TiDB_USERNAME="Username for TiDB" \ --env QDRANT_API_URL="URL for Qdrant vector store (e.g., http://localhost:6333)" \ --env QDRANT_COLLECTION="Name of the vector collection"
How to use
The Cardea Agentic Search MCP server provides an agentic search capability that can perform vector-based searches using Qdrant, keyword-based searches using TiDB, or a combination of both. The server exposes a single tool named ‘search’ which accepts a query string and returns a list of results. You can leverage it to retrieve results that are semantically relevant (vector search), exact-match keywords (keyword search), or a hybrid of both depending on how you configure the underlying data sources. To use it, connect to the MCP server via your preferred MCP client, issue the search tool with your query, and process the returned results.
Typical usage involves configuring the backend connections to Qdrant for vector search and TiDB for keyword search, then issuing a search query. The tool will route the query to the appropriate backends and merge results as needed, returning a structured list of results suitable for downstream agents or applications.
How to install
Prerequisites:
- Node.js (LTS) and npm installed
- Access to a Qdrant instance and a TiDB instance (or endpoints for your environment)
Installation steps:
-
Ensure Node.js and npm are installed on your system. Verify with: node -v npm -v
-
Install and run the MCP server using npx (this pulls the package and runs it in-place):
npx -y cardea-mcp-agentic-search
- If you prefer to install locally for development, you can install the package and run the server script (adjust command if the repository provides a different start script):
npm install cardea-mcp-agentic-search
node node_modules/cardea-mcp-agentic-search/server.js
- Configure environment variables (examples):
export QDRANT_API_URL=http://localhost:6333
export QDRANT_COLLECTION=agentic_search
export TiDB_DSN=root:password@tcp(127.0.0.1:4000)/
export TiDB_DATABASE=agentic
export TiDB_USERNAME=root
export TiDB_PASSWORD=password
- Start the service with the configured environment. The exact startup command may vary depending on how the package exposes the entry point. Refer to the package’s documentation for the exact start command.
Additional notes
Notes and tips:
- Ensure Qdrant is reachable and the specified collection exists before starting the server.
- Ensure TiDB is reachable and the DSN/database credentials are correct.
- The MCP server may support both vector and keyword search depending on backend configuration; tune Qdrant and TiDB connections as needed.
- If you encounter authentication or network errors, verify network ACLs and firewall rules between the MCP host and Qdrant/TiDB endpoints.
- If the provider package names or entry points differ, consult the repository’s DEVELOPMENT or USAGE docs for exact startup instructions.
Related MCP Servers
PPTAgent
An Agentic Framework for Reflective PowerPoint Generation
boltmcp
MCP server from boltmcp/boltmcp
cursor-rust-tools
A MCP server to allow the LLM in Cursor to access Rust Analyzer, Crate Docs and Cargo Commands.
srag
Semantic code search and RAG system written in Rust with tree-sitter chunking, MCP server for IDE integration, prompt injection detection, and secret redaction
repo-stargazer
Talk to your starred github repositories
docrag
AI-powered documentation RAG system with MCP server for Claude Code. Search and retrieve technical documentation on-demand with vector embeddings and smart web scraping.