Best MCP servers for PostgreSQL in 2026

TL;DR: MCP servers for PostgreSQL fall into two groups: provisioning servers that create and manage databases, and data-access servers that query them. The Aiven MCP server is in the first group (with bounded read-only SQL); Supabase and Neon MCP servers span both; Postgres MCP Pro, DBHub and pgEdge are mainly about working with an existing database. All details are [TO VERIFY].

Read this before the table

Lists like this often mix tools that do different jobs. Two categories matter.

• Provisioning and management: create a database, list plans, change settings.
• Data access: list tables, run SQL, explain query plans.

The Aiven MCP server belongs to the first category. It provisions and manages Aiven services and runs bounded read-only SQL. It is not a general data-access MCP, and we say that because it is the likeliest thing to be misrepresented on a page like this one, written by Aiven.

Comparison

Which one "wins" depends on your task, so we do not rank them. Third-party lists such as bytebase.com and glama.ai/mcp/best/postgresql take different approaches; read them too.

How they fit together

flowchart LR
  AG["Agent: Claude Code, Cursor, Codex, Gemini CLI"] --> P["Provisioning MCP: Aiven, Neon, Supabase"]
  AG --> D["Data-access MCP: Postgres MCP Pro, DBHub, pgEdge"]
  P --> DB[("PostgreSQL")]
  D --> DB

A common pattern: the agent uses a provisioning server to create the database, then a data-access server with a restricted role to work inside it.

Copy-paste setup

Most clients accept a JSON block like the one below. Field names differ between clients, and the package or command for each server is [TO VERIFY]; take it from the server's own README.

{
  "mcpServers": {
    "aiven": {
      "command": "<aiven-mcp-command>",
      "args": ["<args-from-aiven-docs>"],
      "env": { "AIVEN_TOKEN": "${AIVEN_TOKEN}" }
    },
    "postgres": {
      "command": "<postgres-mcp-command>",
      "args": ["<args-from-server-readme>"],
      "env": { "DATABASE_URL": "${DATABASE_URL}" }
    }
  }
}

For Claude Code, adding a server is typically done with claude mcp add, and the exact arguments depend on the server. [TO VERIFY]

Create a read-only role first so the data-access server cannot change anything:

create role agent_ro login password '<set-from-secret-manager>';
grant connect on database defaultdb to agent_ro;
grant usage on schema public to agent_ro;
grant select on all tables in schema public to agent_ro;
alter default privileges in schema public grant select on tables to agent_ro;

Then connect with verification on:

export DATABASE_URL="postgres://agent_ro:<password>@<host>:<port>/defaultdb?sslmode=verify-full&sslrootcert=./ca.pem"
psql "$DATABASE_URL" -c "select current_user;"

Choosing by task

For a worked example of the Aiven side, see Aiven MCP for PostgreSQL: create, connect, query and Run a PostgreSQL MCP server on Aiven Runtime.

Decision guide

Choose Postgres MCP Pro or DBHub if you already have a database and want an agent to inspect or query it.

Choose the Supabase or Neon MCP server if you are on those platforms and want one server for project management and SQL.

Choose Aiven if you want an agent to provision and manage services (PostgreSQL, Kafka, ClickHouse) on Aiven, and you will use a separate client or data-access server for the data itself.

Verification note

Dated 2026-11-09. AI-generated and unreviewed. Every server's capabilities, install commands, hosting model and permission modes are unverified, and competitor facts in particular change quickly. Re-verify against each project's repository before relying on this table.