memvid-mcp-rs

# Memvid MCP Server Unofficial Model Context Protocol (MCP) server interface to [Memvid](https://github.com/memvid/memvid) - a video-frame memory system for AI agents. ## What is Memvid? Memvid is a memory layer for AI agents that stores knowledge as compressed "Smart Frames" in a single `.mv2` file (similar to how video stores frames). This approach provides: - **10x more storage efficiency** than traditional vector databases - **Sub-5ms memory access times** - **Append-only, crash-safe storage** with time-travel capabilities - **Single portable file** containing data, embeddings, search structure, and metadata - **No external database or infrastructure required** ## Features - **Persistent long-term memory** for AI agents - **Semantic search** with natural language queries - **Instant retrieval** from a single portable file - **Serverless architecture** - no database infrastructure needed - **Model-agnostic** - works with any LLM that supports MCP ## Installation ### From Cargo (Rust) ```bash cargo install --git https://github.com/mystique09/memvid-mcp.git ``` ### From Source ```bash git clone https://github.com/mystique09/memvid-mcp.git cd memvid-mcp cargo build --release # The binary will be at target/release/memvid-mcp.exe (Windows) or target/release/memvid-mcp (Linux/Mac) ``` ## Configuration ### Continue Add to your `mcpServers/new-mcp-server.yaml`: ```yaml mcpServers: - name: Memvid MCP type: stdio command: memvid-mcp ``` ### Claude Desktop Add to your Claude Desktop config (`claude_desktop_config.json`): ```json { "mcpServers": { "memvid": { "command": "memvid-mcp", "env": {} } } } ``` ### Cline (VS Code Extension) Add to your Cline settings: ```json { "mcpServers": [ { "name": "memvid", "command": "memvid-mcp" } ] } ``` ## Available Tools | Tool | Purpose | Description | |-------|----------|-------------| | `ping` | Health check | Returns "pong" to verify server is running | | `add_chunks` | Store memory | Stores text chunks as memory frames with semantic embeddings. Returns count and sequence IDs | | `search` | Query memory | Searches using natural language. Returns relevant chunks ranked by semantic similarity | ## Usage Examples ### Store Chunks ``` Use the add_chunks tool to store information: Add chunks ["Meeting discussed Q1 planning", "Action item: Review API integration", "Decision: Use Memvid for memory"] ``` ### Search Memory ``` Use the search tool with natural language queries: Search for "Q1 planning" with top_k=5 and snippet_chars=200 ``` ### Health Check ``` Ping to verify server is responsive: Ping ``` ## Storage All data is stored in `memvid.mv2` in the current working directory. This is a single, portable file that includes: - Compressed frames - Full-text index - Vector index - Timeline metadata The `.mv2` file can be shared, versioned, and backed up like any other file. ## Development ### Build ```bash cargo build ``` ### Run (Development) ```bash cargo run ``` ### Test with MCP Client ```bash # Create test requests echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{"tools":{}},"clientInfo":{"name":"test","version":"1.0.0"}}}' | cargo run ``` ## Requirements - Rust 1.85.0+ - `memvid-core` crate (included as dependency) ## Learn More - [Memvid Documentation](https://docs.memvid.com) - [Memvid GitHub](https://github.com/memvid/memvid) - [Model Context Protocol Spec](https://modelcontextprotocol.io) ## License Apache License 2.0 ## Contributing Contributions welcome! Please feel free to submit a Pull Request.