Tessera Brings Persistent Memory and Local RAG Search to Claude Desktop via MCP
Key Takeaways
- ▸Tessera enables Claude Desktop to search across entire local document collections and maintain memory between conversation sessions
- ▸The system operates entirely locally with zero external dependencies, requiring no API keys or cloud services
- ▸Uses fastembed and LanceDB for efficient local vector search with incremental indexing of only changed files
Summary
An open-source project called Tessera has emerged to address a significant limitation in Claude Desktop: the inability to search across multiple documents and maintain memory between sessions. Developed by the besslframework-stack team, Tessera functions as a Model Context Protocol (MCP) server that indexes local documents into a vector store, enabling Claude to automatically search through entire workspaces and remember context across conversations.
The system operates entirely locally without requiring external dependencies like Ollama, Docker, or API keys. It uses fastembed (ONNX) and LanceDB for document indexing and supports various file formats including Markdown, CSV, and session logs. Tessera only re-indexes changed files during sync operations, making it efficient for ongoing use. The tool also includes a knowledge graph feature to visualize connections between documents.
Installation is streamlined through a simple setup process that guides users through selecting document directories, choosing which folders to index, and automatically generating the necessary configuration. The embedding model download is a one-time operation of approximately 220MB. All processing and data storage remain on the user's machine, addressing privacy concerns about sending documents to external services.
- Includes a knowledge graph visualization feature to show document relationships
- Open-sourced under AGPL-3.0 license with streamlined setup via 'tessera init' command
Editorial Opinion
Tessera represents an important step in making AI assistants more practically useful for knowledge workers who deal with large document collections. The focus on local-first operation addresses legitimate privacy concerns while the MCP integration shows how standardized protocols can enable powerful third-party enhancements. However, the project's maturity and long-term maintenance remain to be seen, as community-driven tools often face sustainability challenges.
