Model Context Protocol ( MCP ) Python server to use with continue.dev

What is Model Context Protocol ( MCP ) Python server to use with continue.dev?

This MCP server provides access to an Obsidian.md vault with search and read tools, as well as a web deep research tool similar to Perplexity.ai. It is designed for AI agents and knowledge retrieval, and can be used as a shared knowledge base and memory across projects.

How to use Model Context Protocol ( MCP ) Python server to use with continue.dev?

Clone the repository, create a .env file from env.example, then run uv run --project <local copy> mcps --vault <Vault Folder> --reindex to index the vault. Start the server with uv run --project <local copy> mcps --vault <Vault Folder> --port 1234. It runs as an HTTP MCP server.

Key features of Model Context Protocol ( MCP ) Python server to use with continue.dev

• Web deep research agent using iterative search and content extraction
• Hybrid vector + BM25 search for Obsidian vault notes
• Note chunking by headers with frontmatter metadata (title, description, tags)
• Summary chunk generated by LLM for improved recall
• Reranking via Cohere/Voiage.ai API or Reciprocal Rank Fusion
• Tag and file path filtering for narrowed searches

Use cases of Model Context Protocol ( MCP ) Python server to use with continue.dev

• AI coding agents (Claude Code, Cursor) retrieving relevant notes from a large Obsidian vault
• Answering technical or academic questions via web research, then storing results in the vault
• Shared knowledge base accessible by multiple AI tools across different projects

FAQ from Model Context Protocol ( MCP ) Python server to use with continue.dev

What LLM provider does the server require?

It uses a single LLM API provider. The author uses LiteLLM Gateway, but Openrouter also works.

Where are the indexed notes and embeddings stored?

Indexed chunks, metadata, and vector embeddings are stored in a local database. The original vault remains in its folder.

What transport does the server use?

The server runs as an HTTP MCP server on a configurable port (e.g., 1234). This allows a single instance to be shared across multiple AI tools.

What are the limitations of the web research tool?

It does not attempt to bypass bot protections, so some public sources may be inaccessible. It is optimized for technical/academic content.

How does the search handle note format?

The server assumes a specific note structure: frontmatter with title, description, tags, content split by headers, wikilinks, and notes no longer than 200 lines. It works best with vaults following this convention.