Self-hosted TiDB MCP alternatives
PingCAP's TiDB MCP server installs locally and runs over stdio, pointing at a TiDB deployment you operate. The process and connection details stay on your own machine, which fits a team running its own distributed SQL cluster. If you want the same local server pointed at a different engine, the picks below all run that way.
A caveat that applies across the list: running the MCP server locally controls where the process and credentials live, but queries still reach whichever database you connect to. Self-host that database too and the data stays on your hardware.
The 8 best self-hosted alternatives
The lightest option by far, the SQLite server through Bytebase DBHub runs an agent against a local database file with execute_sql and search_objects. No cluster, just a file on disk.
Set up SQLite (DBHub) →DBHub connects an agent to a self-managed Postgres via a DSN over stdio, the Postgres-side counterpart to TiDB's MySQL dialect. It runs SQL and searches objects from a process you control.
Set up DBHub (Postgres) →The community Supabase server runs locally and connects to a Supabase project: run SQL, inspect schema, read logs, and manage edge functions. It pairs Postgres with an application platform rather than distributed scale.
Set up Supabase →MongoDB's official server installs locally and queries and manages databases with find, aggregate, and insert, plus Atlas administration. The document specialist for data that does not fit relational tables.
Set up MongoDB →Pointed at a self-hosted cluster, ClickHouse's official server lists databases and tables and runs read-only SQL. Its column-oriented design handles analytics that a transactional engine struggles with.
Set up ClickHouse →Elastic's official server runs locally and lists indices, reads mappings, and runs full-text and ES|QL queries against a cluster you operate. The search engine to reach for when SQL filtering is not enough.
Set up Elasticsearch →For in-memory state on your own infrastructure, Redis's official server reads and writes strings, hashes, lists, streams, JSON, and vector search over stdio. It complements a primary store rather than replacing it.
Set up Redis →When relationships matter more than rows, Neo4j's official server introspects a graph schema and runs read or write Cypher against your own deployment. Connected data and traversals are its strength over relational SQL.
Set up Neo4j →
How to choose
All of these run as local stdio servers, so the choice is the engine, not the hosting. TiDB gives MySQL-compatible distributed scale; if you do not need that, DBHub for Postgres or SQLite keeps it simple. ClickHouse handles analytics, Elasticsearch handles search, MongoDB handles documents, Neo4j handles graphs, and Redis handles caching. Self-host the database too and both the server and the data stay on your machine.
FAQ
- Can the TiDB MCP server be self-hosted?
- Yes. The server installs and runs locally over stdio, and TiDB itself can run on your own hardware, so both the MCP process and the database can stay on your infrastructure.
- Does running a database MCP server locally keep the data on my machine?
- It keeps the MCP server process and connection details local. Whether the data stays local depends on the database. Self-host Postgres, ClickHouse, Neo4j, or TiDB itself and the data is on your hardware; point at a managed cloud and queries travel there.