Dolt is the database for agents. I’ve been talking about this since March 2025. The three pillars of agentic workflows are a capable model, version control, and tests. Dolt provides version control on your database, the storage layer for almost everything in your enterprise. Use Dolt if you want to bring the success of agentic coding to other places in your enterprise. Agents use the Model Context Protocol (MCP). Is Dolt the MCP Database?

Traditionally, we’ve tried to rank organically on Google through our So you want… series. Over the past year, we have spent a bit of money on advertising, targeting a handful of AI search queries. A well-performing search query for us is “MCP database”. I was a bit baffled by this so I checked out the organic results. The results were lackluster, led by mcpdb.com, which definitely has nothing to do with Model Context Protocol. Big win for Marvel Crisis Protocol. This article is my attempt to rectify that injustice.

What is an MCP Database? This article sets out to answer that question.

What is MCP?#

MCP stands for Model Context Protocol. MCP was introduced by Anthropic very early in the Claude agentic coding revolution. MCP defines a simple HTTP-based standard for exposing tools to Claude and, now, other agents. Tools are actions an agent can take. Book this reservation. Search the web for this. Run this SQL query. MCP became all the rage in Q2 2025 as Claude finally crossed a critical usability threshold. Every service got an MCP server so it could work better with Claude.

However, MCP was mostly a 2025 concept. Over the back half of 2025, Claude and other agents became very good at operating the command-line interfaces (CLIs) humans use, obviating the need for a special agent interface. CLIs are critical for agentic coding so the model companies put a lot of effort into making their agents good at using them. Thus, MCP servers had their moment but modern agents seem to prefer the same CLIs humans use.

What is a Database?#

A database is a repository of structured data. A database can refer to the data inside the database (e.g., the Internet Movie Database, IMDb) or the actual tool storing the data (e.g., MySQL). Carnegie Mellon’s Database of Databases illustrates both meanings.

What is an MCP Database?#

So when a person is searching for MCP database, do they mean a queryable list of MCP servers? Or do they mean the MCP/database combination that works best?

Queryable List#

If you’re looking for a queryable list of MCP servers, Anthropic hosts a GitHub repository with reference implementations. That repository links to an MCP Registry. The problem with this registry is you can only search by name. I found a couple of competing registries that let you filter by category: MCPMarket and MCPServers.org.

If you’re looking for an MCP server, I bet the tool you want to connect to has one. It might be better to just search for the tool’s name followed by “mcp server”. Here are the results for “dolt mcp server”. I’m not certain why anyone would want the browsing experience of a registry in this case.

The Best Database for MCP#

So, I’m pretty sure people searching for “MCP database” want to know which database works best with an MCP server. Here, I have some opinions. What should you look for in an MCP database?

1. Known API#

You should choose a database with a known API. For SQL databases, I would stick to the standard open source flavors: Postgres, MySQL, or SQLite. I would not choose an MCP database with a custom API.

2. Well Tested#

Again, stick to the well-trodden path. You want an MCP server and database from a reputable source. Prefer the database’s own supported MCP server over a third-party variant.

3. Easy to Deploy#

This is where you will start to see some differences among MCP offerings. The standard way to run MCP servers is as a separate process that acts as a bridge between your database and agents. I would prefer an MCP server that shipped with the database itself, not as a separate service. Then, you only need to run one server that serves standard clients and MCP clients, usually on different ports.

4. Write Safety#

This is the big one. Agents are known for making catastrophic mistakes, especially with databases. You are going to want to adopt a database that can restrict or isolate MCP-generated operations. Ideally, you want a database that allows your MCP server to support safe agentic writes.

The first method to do this is permissions. Restrict the permissions of the MCP server’s database user, or disable write operations entirely. The supported Postgres MCP server shipped without write access initially. Most databases will support this method of write safety.

The better way to do this is to create a write sandbox. A new generation of version-controlled databases exists that allows you to create branches for write isolation. You should pick one of these for your MCP database.

Let’s walk through them.

Neon#

Tagline

Serverless Postgres

Initial Release

June 2021

GitHub

https://github.com/neondatabase/neon

SQL Flavor

Postgres

Neon is a very popular open-source Postgres variant. Neon separates storage and compute and substitutes the PostgreSQL storage layer by redistributing data across a cluster of nodes on a copy-on-write filesystem. Neon is the default database for the popular App Builders, Replit and Vercel.

Neon supports branches using Postgres’ copy-on-write functionality. This method does not allow for diff or merge so these “branches” are more aptly called forks.

Neon has an official MCP server that supports write operations on branches. However, the MCP server is a separate product. This is my third choice for an MCP database.

Supabase#

Tagline

The Postgres development platform.

Initial Release

May 2020

GitHub

https://github.com/supabase/supabase

SQL Flavor

Postgres

Supabase is the fastest growing database on the planet. To say it’s a database is a bit of an understatement. It starts as Postgres and then layers on a bunch of best-of-breed open source Postgres tools to give you more of a “backend in a box”. One of the included tools is an MCP server. Like Neon, Supabase also supports Postgres’ copy-on-write “branches”… er, forks.

Because of the built-in MCP server, I would prefer this solution to Neon, making Supabase my second choice as an MCP database.

Dolt#

Tagline

Git for Data

Initial Release

August 2019

GitHub

https://github.com/dolthub/dolt

SQL Flavor

MySQL, Postgres, SQLite

I mentioned Dolt in the opening. I’m the CEO of the company that built it so I’m a bit biased.

Dolt is the world’s only truly version-controlled database. All the version control functionality you know and love from Git is available in Dolt, all with the query performance of MySQL or Postgres. Version control is essential for agentic workloads because agents make mistakes. Give your agent a branch or clone and let them run wild. Merge only if the change is good, just like you are used to with coding agents.

Dolt ships with a built-in MCP server and full version control delivers best-of-breed write safety. Dolt is my first choice for an MCP database.

Interested in learning more? Come by our Discord and I’d be happy to give you a demo.