# Destrayon/Connapse [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/Destrayon/Connapse  
**GitHub Stars:** 20  
**Views:** 5  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/destrayon-connapse

## Description
⃣ 🏠 🍎 🪟 🐧 - Self-hosted knowledge backend for AI agents with hybrid vector + keyword search, container-isolated indexes, and 11 MCP tools. .NET, Docker-ready.

## Claude Desktop Quick Installation
Remote MCP endpoint (confidence: high). Install path detected from listing signals. Add as a URL/SSE server in your client:

```json
"mcpServers": {
  "connapse": {
    "url": "http://localhost:5001/mcp"
  }
}
```

## Documentation

## What Destrayon/Connapse MCP server does

Destrayon/Connapse MCP server gives agents access to a self-hosted knowledge backend organized into containers. Agents can create containers, add documents, inspect stored content, search indexed knowledge, and remove files or containers through MCP. The backend is intended for persistent context across conversations rather than temporary prompt attachments.

Connapse accepts uploaded documents and can use local filesystems or existing Amazon S3 buckets as sources. External sources are mirrored read-only: agents can search them, but cannot browse or modify them through container file operations. The project is categorized under 🧠 Knowledge & Memory and is built with .NET 10.

## How it works

The MCP endpoint is served by the Connapse web application at `/mcp` using streamable HTTP. An agent authenticates with an agent API key passed in the `X-Agent-Api-Key` header. The application stores metadata in PostgreSQL with the pgvector extension and uses MinIO for object storage when deployed through the supplied Docker Compose setup.

Search can use semantic matching, keyword matching, or a hybrid of both within a selected container. Uploaded files are parsed and indexed so that search results can provide source citations. Container statistics expose document, chunk, storage, and embedding counts. Embedding providers, chunking parameters, and search settings can be changed at runtime without restarting the application. Ollama is available as an optional local embedding provider.

## Setup and configuration

The documented quick start requires Docker and Docker Compose. Clone the repository, set `CONNAPSE_ADMIN_EMAIL`, `CONNAPSE_ADMIN_PASSWORD`, and an `Identity__Jwt__Secret`, then run `docker-compose up -d`. The first startup initializes PostgreSQL, applies Entity Framework Core migrations, creates MinIO buckets, seeds the administrator account, and starts the web application on port 5001.

Create an agent API key in the web interface under Settings → Agent API Keys. Configure the MCP URL and API key in Claude Desktop, Claude Code, VS Code, or Cursor. Development can run PostgreSQL and MinIO through Docker while launching `src/Connapse.Web` with the .NET SDK.

## Tools and capabilities

Destrayon/Connapse MCP server exposes these 11 tools:

- Create, list, delete, and inspect containers with `container_create`, `container_list`, `container_delete`, and `container_stats`.
- Upload one file or up to 100 files with `upload_file` and `bulk_upload`.
- Browse paths with `list_files` and retrieve parsed document text with `get_document`.
- Remove one file or up to 100 files with `delete_file` and `bulk_delete`.
- Search a container using `search_knowledge` with semantic, keyword, or hybrid modes.

## Limitations and notes

The MCP service requires a running Connapse deployment and an agent API key. Source containers are searchable but are not writable or browsable through the file and write tools. Docker is the documented deployment path, while local development requires the .NET 10 SDK. The separate `connapse` command-line client is published as its own project and is not an MCP tool.

_Full upstream README: https://allmcps.com/mcp/destrayon-connapse/readme_