Connapse
Run a self-hosted knowledge backend with hybrid search so your agent can store and retrieve project memory through MCP instead of losing context every session.
Overview
Connapse is an MCP server for the Build phase that provides a self-hosted knowledge backend with hybrid search and MCP tools for AI agents.
What is this MCP server?
- Self-hosted knowledge backend designed for AI agents (Connapse v0.3.2)
- Hybrid search over your corpus—not only keyword or only vector
- MCP tools on streamable-http at localhost:5001/mcp by default
- OCI package ghcr.io/destrayon/connapse for repeatable deploys
- Open-source repository on GitHub (Destrayon/Connapse)
- Catalog version 0.3.2
- Default MCP transport URL http://localhost:5001/mcp
- OCI identifier ghcr.io/destrayon/connapse:v0.3.2
Community signal: 13 GitHub stars.
What problem does it solve?
Agent sessions forget prior decisions and docs unless you rebuild context manually or pay for opaque cloud memory with weak hybrid retrieval.
Who is it for?
Indie builders running local or VPS stacks who want agent-accessible RAG with hybrid search without shipping source code to a third-party memory SaaS.
Skip if: Builders who want zero ops, fully managed vector DB, or retailer-style remote MCP with no self-hosting.
What do I get? / Deliverables
After you deploy Connapse and register its MCP endpoint, your agent can query and update a persistent knowledge store on hardware you control.
- Running Connapse instance with MCP tools for ingest and hybrid search
- Agent-retrievable knowledge store under your control
Recommended MCP Servers
Journey fit
Building agent-powered products means wiring durable retrieval; Connapse sits in Build as the persistence and search layer behind MCP tools. Agent-tooling is where you add memory, RAG, and tool backends; hybrid search plus MCP exposure is exactly that integration shelf.
How it compares
Self-hosted knowledge and hybrid-search MCP backend, not a single-purpose IDE skill or hosted deals API.
Common Questions / FAQ
Who is Connapse for?
Solo developers and small teams building agent workflows who need self-hosted persistence and hybrid search exposed through standard MCP tools.
When should I use Connapse?
Use it while building agent-tooling or integrations when project knowledge must survive restarts and be retrieved with hybrid search from Claude Code, Cursor, or similar clients.
How do I add Connapse to my agent?
Run the Connapse service (e.g. OCI image ghcr.io/destrayon/connapse:v0.3.2), ensure MCP is on http://localhost:5001/mcp, and register that streamable-http URL in your agent MCP config.