Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Turbovec MCP Server search my memory for how we handle user authentication"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
A local-first Model Context Protocol server that gives your AI assistant a persistent memory it can search, traverse and extend across sessions.
Everything runs in-process on your machine: a SQLite knowledge graph, SQLite FTS5 for keyword search, and the turbovec vector index for semantic search. No network calls, no external database.
Works with Claude Desktop, Cursor, Cline / Roo Code, Windsurf, Zoo Code, and any other MCP client.
Problem | What Turbovec does |
Pasting whole files into chat burns tokens and degrades reasoning | Stores them once, returns only the relevant chunks |
Closing a chat loses every decision and convention | Persists entities, relationships and session summaries to disk |
Keyword search misses anything phrased differently | Runs vector and keyword search, then fuses the rankings |
Saved notes drift into an orphaned, stale pile | A background librarian clusters, links and prunes them |
Related MCP server: Heimdall MCP Server
A query runs through two independent channels. Their rankings are merged with Reciprocal Rank Fusion, so no score calibration between them is needed.
graph LR
Q[Query] --> V["Vector channel<br/>(turbovec ANN)"]
Q --> L["Lexical channel<br/>(SQLite FTS5)"]
V --> R{{"Reciprocal<br/>Rank Fusion"}}
L --> R
R --> H["Hydrate nodes<br/>+ observations"]
H --> X["Cross-encoder rerank<br/>(optional)"]
X --> O[Ranked results]Stored alongside the text is a typed graph — entities, weighted relationships and timestamped observations — so results can be expanded into their surrounding context instead of returned as isolated snippets.
git clone https://github.com/henny-bee/Turbovec-MCP-Server.git
cd turbovec-mcp-server
python -m venv venv
source venv/bin/activate # Windows: .\venv\Scripts\activate
pip install -r requirements.txt
python main.py # startup is logged to server.logThen point your MCP client at it — for Claude Desktop, in claude_desktop_config.json:
{
"mcpServers": {
"turbovec": {
"command": "/absolute/path/to/venv/bin/python",
"args": ["/absolute/path/to/main.py"],
"env": { "PYTHONUNBUFFERED": "1" }
}
}
}Use the absolute path to the venv's Python, not bare python — otherwise your
editor runs the system interpreter and the import fails. Setup for other editors,
Docker and SSE mode is in the Configuration Guide.
Guide | Contents |
Editor integration, environment variables, Docker, dashboard, storage schema | |
All 35 MCP tools and 2 prompts, by category | |
Rules that make your assistant use the memory unprompted | |
Module map, design rules, performance decisions |
pip install -r requirements-dev.txt
pytest # 78 unit and integration tests
black . # formatting, enforced in CITests cover graph consistency, transactional rollback, hybrid search ranking, rank fusion, embedding cache and batching, index persistence, clustering, semantic radar, and the librarian cycle.
MIT — see LICENSE.
Sponsored by
This server cannot be deployed
Shared memory for coding agents. Stop re-explaining your codebase every session.
Persistent memory for AI agents. Semantic search, memory graph, W3C DID identity.
Hosted persistent memory with semantic search, importance and TTL for AI agents.
Universal memory for AI agents and tools. Save, organize and search context anywhere.
