Put your team
and its AI agents
on the same page.
Claude, Codex, ChatGPT, and Gemini read and update the same records through MCP — backed by the Postgres running in this Docker compose.
Use Hutch to
Run repeatable playbooks
Keep instructions, files, owners, and current status together.
Store and share structured data
Keep research, records, and files where every agent can find and update them.
Hand work between agents
One agent collects, another processes, and the next run sees the result.
$ claude mcp add --transport http hutch-core http://localhost:3000/api/mcp $ claude > "store this in my reading-list collection" Created reading-list. Saved 1 record.
The work stops living in separate chats
A useful answer in one session should become something every agent can use next time. Save the result, the source data, or the improved process to Hutch, and every run starts from the latest version.
Records render however the data shapes up
Views are seeded automatically from the inferred schema — a select field gets a Kanban, a date field gets a Calendar.
Table
Kanban
Calendar
Timeline
Gallery
Use Hutch from any MCP-aware client
Everyone on the team connects their own; this Postgres sits behind all of them.
What is Hutch Core?
The open-source, self-hosted core of Hutch — a headless MCP server backed by Postgres. Schema-optional collections, JSONB records, and views, reachable over MCP and REST.
Do I need an API key?
Not for local/self-hosted use. Set HUTCH_API_KEY to require bearer auth; leave it unset and this instance runs single-user, anonymous, trusted-localhost.
Is this a black box?
No. It's Postgres underneath, the core is open source (AGPL), and everything is reachable over plain MCP and REST.
Create your first collection
Connect an AI agent and say “save to hutch” — or open the dashboard to look around.