HutchDB: a database for AI agents
HutchDB
HutchDB MCP server: shared agent records in Postgres, schema-free collections, search and filters for Claude Code and Cursor
- 26 ★ on the repository
- Last commit
- AGPL-3.0
Extended access
Why this level
- Without HUTCH_API_KEY the server trusts every request, so data on a public host is open to anyone
- The agent writes, edits and deletes records in your Postgres database
Install
- Works in
Manual install
docker compose upRun it in the project folder after git clone https://github.com/ExpeditedProjects/hutchdb, then add the server to the MCP config.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Without HUTCH_API_KEY the server trusts every request, so data on a public host is open to anyone
- The agent writes, edits and deletes records in your Postgres database
This is third-party code. Review the repository files before installing.
What it does
HutchDB Core is a headless, single-user MCP server on a Postgres database you control. Claude Code, Codex, Cursor and VS Code agents read and update the same collections across sessions. Records are stored as arbitrary JSON in JSONB. A collection is created on the first write and a schema is optional. Queries use Mongo-style filter operators ($gt, $in, $exists, $contains) and full-text search. Tools prefixed hutch_ cover collections, records, schema and views, with import and export as CSV and JSON. Typical uses are accumulating research, trackers in place of spreadsheets, pipelines where one agent collects data and another processes it by a status field, and working memory for a long project. The HutchDB Cloud edition adds a web interface, OAuth and organizations, which the core lacks.
Who it is for. Developers and analysts whose agents accumulate data across sessions and who have outgrown a TODO.md or Markdown notes.
Good fit when
- Several agents update one list and must not overwrite each other
- Research grows session after session and later you need to ask what changed since last month
- Bugs, requests or a content calendar are easier as records so the agent can filter the open ones by severity
- One agent collects leads and another enriches them, with a record status marking the pipeline stage
- Project decisions and open questions have to outlive a closed context window
Not a fit when
- You need a web interface, published pages or social login: those exist only in HutchDB Cloud
- You want to connect the server to claude.ai in the browser: the core ships no OAuth server and the web client requires one
- The data is far beyond an agent notebook and needs a strict schema and complex joins: this is not a relational DBMS
Example request
Save these five competitor pricing pages to the competitor-pricing collection and tell me who changed prices since last monthMCP
- Transport
- http
- Authentication
- key is optional
Environment variables
HUTCH_API_KEYsecret- Static key: when set, every MCP call needs a Bearer token. Without it all requests are trusted.
HUTCH_DATABASE_URLsecret- Postgres connection string when running without Docker.
Limitations
The core is single-user and has no dashboard. Without the HUTCH_API_KEY variable the server trusts every request: fine for localhost, not for a public host. Running it takes Docker, or Node with Postgres and the HUTCH_DATABASE_URL variable.
The license is AGPL v3. The authors say it is stable enough for a personal self-host, and breaking changes land straight on main, so pin a commit for stability. Claude Code, Cursor, Codex and VS Code work with a static token, while claude.ai on the web cannot connect without OAuth.
How to disable. Remove the hutch entry from your MCP config and stop the container with docker compose down in the project folder. The data stays in your Postgres database and is deleted separately.
FAQ
How do I run HutchDB for Claude Code?
Clone the repository and run docker compose up; the core comes up on port 3000. Add a server to the MCP config with type http, the address http://localhost:3000/api/mcp and an Authorization header with your token.
Do I need to describe a collection schema?
No. A collection is created on first write and a record can be any JSON. You can describe a schema later and turn on validation; the schema tools can infer one from the data.
How does HutchDB Core differ from HutchDB Cloud?
The core is open, single-user and has no interface. The cloud adds a web dashboard, published pages, OAuth 2.1, social login and organizations, and it works with claude.ai on the web.
How do I protect the server on a public host?
Set HUTCH_API_KEY, for example with openssl rand -hex 32, before starting. Then every MCP call must carry a Bearer token. Without a key the server trusts everyone.
How do I move data?
Collections export and import as CSV and JSON through the record tools.
Discussion
No questions or reviews yet.
Related
MCP serversA token-lean MCP server for PostgreSQL, MySQL, MariaDB, SQL Server and SQLite with a read-only mode
A lightweight database workspace with an MCP server: the agent reads the schema, runs queries and explores data in PostgreSQL, MySQL and more
A Rust terminal database client with MCP and a JSON CLI: agents read schemas and run queries within profile permissions.
One MCP server for SQLite, SQL Server, PostgreSQL and MySQL: read and change data, edit schemas and export to CSV and JSON
A database CLI for dozens of engines with an agent mode: JSON output, a read-only guard, limits and a broker that hides passwords
A DataHub MCP server: an agent searches tables and columns by metadata, traces lineage and finds analysts' SQL queries
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| HutchDB: a database for AI agents (this card) | yes | key is optional | 26 | |
| Supermemory | yes | OAuth | 31,602 | stores memories and context, while HutchDB holds structured records with filters and statuses |
| agentmemory | yes | not required | 30,340 | an agent's memory for its own notes, whereas HutchDB serves as a shared working database for several agents |
README in short
The README explains the purpose, usage patterns with example prompts for the agent, a quick start with and without Docker, a Core versus Cloud comparison and the architecture on Next.js, Drizzle and Postgres. Contributors sign a CLA so the maintainer can dual-license the code.