Integrations
Hydrate ships in two integration shapes. Native hooks on Claude Code, MCP everywhere else.
Two shapes, one store
Whichever path your agent takes, it reads from and writes to the same Hydrate fact store. The shape of the integration just decides how the agent reaches it.
| Shape | Tools | Behaviour |
|---|---|---|
| Native hooks | Claude Code |
Zero-friction. Fires on every turn automatically. The
UserPromptSubmit hook injects facts; the Stop
hook extracts them. You never call a tool by hand.
|
| MCP | Gemini CLI, Cursor, Cline, Continue, Windsurf, Zed, Claude Desktop |
The model chooses to call Hydrate tools. One binary
(hydrate-mcp) speaks MCP over stdio and every MCP-capable
client gets hydrate_recall,
hydrate_save_fact, and hydrate_list_projects.
|
Status
- Claude Code — ✅ shipped. Setup docs at /docs/install.
- Gemini CLI — ✅ shipped today. Setup docs at /docs/integrations/gemini-cli.
- Cursor — coming soon. MCP pattern is identical to Gemini.
- Cline — coming soon.
- Continue.dev — coming soon.
- Windsurf — coming soon.
- Zed — coming soon.
- Claude Desktop — coming soon.
The architectural choice
One hydrate-mcp binary, N clients, a documented per-tool config
snippet for each. We didn't write a bespoke plugin for every agent; we shipped
the protocol and let each client point at the same binary.