Waggle Mcp
Give Claude Code and other MCP clients durable, graph-backed conversational memory across sessions instead of re-explaining project context every time.
Overview
Waggle-mcp is an MCP server for the Build phase that stores graph-backed conversational memory agents can reuse across sessions.
What is this MCP server?
- Graph-backed conversational memory exposed via MCP (PyPI package waggle-mcp, v0.1.8)
- Pluggable backends: SQLite default for local solo use or Neo4j for service deployments
- Configurable stdio transport via WAGGLE_TRANSPORT (default stdio)
- Tenant isolation via WAGGLE_DEFAULT_TENANT_ID (default local-default)
- Local embeddings default model all-MiniLM-L6-v2 via sentence-transformers (per manifest env)
- Package version 0.1.8 on PyPI identifier waggle-mcp
- Default backend sqlite; neo4j optional via WAGGLE_BACKEND
- Default embedding model env WAGGLE_MODEL=all-MiniLM-L6-v2
Community signal: 16 GitHub stars.
What problem does it solve?
Coding agents forget prior context between chats, forcing you to repeat architecture decisions and user preferences.
Who is it for?
Builders running multi-day agent workflows who want local SQLite memory first and an upgrade path to Neo4j.
Skip if: Teams that only need ephemeral chat history with no MCP integration or no tolerance for embedding-model local compute.
What do I get? / Deliverables
After installing Waggle-mcp, your agent can read and write persistent graph memory locally or against Neo4j under your chosen tenant.
- On-disk SQLite graph memory at configurable path (default ~/.waggle/memory.db)
- MCP-accessible store/recall of conversational entities and links
- Optional multi-tenant memory separation via WAGGLE_DEFAULT_TENANT_ID
Recommended MCP Servers
Journey fit
Persistent agent memory is Build-phase agent-tooling: it wires infrastructure so coding agents retain facts, relationships, and tenant-scoped history while you implement. Agent-tooling is canonical because Waggle is not a UI feature or ship gate—it is the memory substrate MCP tools read and write during development.
How it compares
Graph memory MCP database layer, not a markdown notes skill or hosted chat product memory.
Common Questions / FAQ
Who is Waggle-mcp for?
Developers using MCP-enabled agents who need durable, structured memory beyond a single conversation window.
When should I use Waggle-mcp?
Use it while building agent-powered tools or long refactor sessions where recalling prior constraints, names, and relationships saves rework.
How do I add Waggle-mcp to my agent?
Install the PyPI package waggle-mcp (v0.1.8), set WAGGLE_BACKEND, WAGGLE_DB_PATH or Neo4j credentials, WAGGLE_DEFAULT_TENANT_ID, and register the stdio MCP server in Claude Code, Cursor, or Codex per your client’s MCP config.