Integrations¶

HydraMem is MCP-native: it speaks the Model Context Protocol, so any MCP-capable AI coding tool can use HydraMem as its long-term memory — 100 % locally, with zero cloud calls by default.

Every integration below boils down to two ingredients:

Run the server — hydramem serve (stdio for desktop/CLI clients, or HTTP for shared/remote use).
Point your tool at it — add HydraMem to the tool's MCP config.

Once connected, your agent gets all 18 MCP tools (search, graph traversal, verification, the remember tool, ingestion, stats…). See the MCP tools reference.

Pick your tool¶

Claude Code

Native MCP via claude mcp add. Plus a retention guide.
Gemini CLI

gemini mcp add + an optional save hook.
Antigravity

Google's agentic IDE — MCP store + lifecycle hooks.
OpenClaw

openclaw mcp set — memory for ClawHub agents.
Cursor

MCP server + Cursor hooks (sessionStart / stop / preCompact).
OpenCode

Local or Anthropic provider + HTTP MCP.
GitHub Copilot

VS Code agent mode via .vscode/mcp.json.
Remote / Team server

One shared HydraMem over HTTP for a whole team.

How "remembering" works in HydraMem (honest contract)¶

HydraMem's memory model is agent-driven (see ADR-0005): the agent decides what is worth keeping and calls the remember tool to persist a verified fact mid-conversation, or ingests your Markdown notes with hydramem ingest.

What this means, stated plainly:

✅ HydraMem ingests Markdown documents and agent-submitted knowledge (remember, ingest_directory_tool, submit_session_extraction).
✅ Every relation is verified (SR-MKG + VoG) before it enters the graph.
❌ HydraMem does not ship a raw transcript-miner that parses each tool's proprietary JSONL session files. "Auto-save" hooks below re-ingest your knowledge directory and nudge the agent to remember — they do not scrape the chat log. If a tool-native transcript miner matters to you, it's tracked as potential future work, not a shipped feature.