PostgreSQL MCP server
Commit Nov 9, 2026Provides AI assistants with structured access to PostgreSQL databases for schema exploration, data profiling, and safe read/write operations.
Also known as ralscha/postgresql-mcp
- Stars
- 0
- Tools
- 24
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Nov 9, 2026
- License
- MIT
What is PostgreSQL MCP?
PostgreSQL provides AI assistants with structured access to PostgreSQL databases for schema exploration, data profiling, and safe read/write operations. It exposes 24 tools, such as search_schema, describe_table, list_table.
Summary generated from the project's README
About PostgreSQL MCP server
Connecting AI assistants to PostgreSQL databases for schema exploration and data management is what this server enables. It provides a structured interface for developers to perform tasks ranging from simple table descriptions to complex relationship analysis and query performance inspection. The server supports both local stdio communication and remote HTTP transport, allowing for flexible integration into various client setups.
Security is managed through a tiered access model controlled by the POSTGRESQL_ACCESS_LEVEL variable. Users can operate in READONLY mode for schema searching and data profiling, DML-RW for inserting or updating rows, or DDL-RW for creating and dropping tables. To prevent accidental data loss, the server includes a mutation confirmation system that shows affected rows before executing writes when POSTGRESQL_REQUIRE_CONFIRMATION is active.
Technical features include query explanation via JSON execution plans, automatic decoding of jsonb data, and authoritative row limits to prevent overwhelming the client. Developers can configure connection details such as host, port, and SSL mode through environment variables, and can secure HTTP endpoints using bearer tokens.
What you can do with it
- 1Analyze database schema structuresAsk the AI to find specific tables or describe the columns and keys within a database.
- 2Optimize slow SQL queriesRequest an execution plan to understand why a specific query is performing poorly.
- 3Profile table statisticsGet a summary of row counts, null values, and data distributions for a specific table.
Written from the project's README. Every use case is checked against a quote from it.
PostgreSQL MCP tools 24
Detected from the READMEsearch_schemadescribe_tablelist_tablelist_databaseslist_environmentsprofile_tableinspect_relationshipsinspect_dependenciesexplain_queryread_datatest_connectionvalidate_environment_config
| search_schema | — | |
| describe_table | — | |
| list_table | — | |
| list_databases | — | |
| list_environments | — | |
| profile_table | — | |
| inspect_relationships | — | |
| inspect_dependencies | — | |
| explain_query | — | |
| read_data | — | |
| test_connection | — | |
| validate_environment_config | — |
Show all 24 toolsShow fewer tools
list_schemaslist_extensionslist_viewslist_triggersshow_create_tabletable_sizeinsert_dataupdate_datadelete_datacreate_tablecreate_indexdrop_table
| list_schemas | — | |
| list_extensions | — | |
| list_views | — | |
| list_triggers | — | |
| show_create_table | — | |
| table_size | — | |
| insert_data | — | |
| update_data | — | |
| delete_data | — | |
| create_table | — | |
| create_index | — | |
| drop_table | — |
Configuration
POSTGRESQL_ACCESS_LEVELPOSTGRESQL_DATABASEPOSTGRESQL_HOSTPOSTGRESQL_HTTP_ADDRPOSTGRESQL_HTTP_PATHPOSTGRESQL_PASSWORDPOSTGRESQL_SSLMODEPOSTGRESQL_TRANSPORTPOSTGRESQL_USER
| Name | Description | |
|---|---|---|
| POSTGRESQL_ACCESS_LEVEL | — | |
| POSTGRESQL_DATABASE | — | |
| POSTGRESQL_HOST | — | |
| POSTGRESQL_HTTP_ADDR | — | |
| POSTGRESQL_HTTP_PATH | — | |
| POSTGRESQL_PASSWORD | — | |
| POSTGRESQL_SSLMODE | — | |
| POSTGRESQL_TRANSPORT | — | |
| POSTGRESQL_USER | — |
Automated checks
Ran Nov 14, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachableralscha/postgresql-mcp is public and reachablegithub.com/ralscha/postgresql-mcp ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-11-09github.com/ralscha/postgresql-mcp ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model24 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHubpostgresql-mcp
An MCP (Model Context Protocol) server that gives AI assistants structured, read-optimized access to PostgreSQL databases. It exposes schema exploration, data profiling, relationship analysis, query explanation, and safe read/write operations through a standardized MCP interface.
Features
- 24 MCP tools covering schema search, table description, data profiling, relationship/dependency inspection, query explanation, connection testing, schema/extension/view/trigger listing, CREATE TABLE DDL generation, table sizing, and tiered write access
- Tiered access model via
POSTGRESQL_ACCESS_LEVEL:READONLY(default),DML-RW(adds insert/update/delete),DDL-RW(adds create/drop table/index) - SQL-safe design with identifier validation, exact named-parameter binding, single-statement checks, and database-enforced read-only transactions
- Authoritative row limits that cap
read_dataresults even when the submitted query contains its own largerLIMIT - Mutation confirmation with preview mode that shows affected rows before executing writes when
POSTGRESQL_REQUIRE_CONFIRMATIONis enabled - Explain plan via
EXPLAIN (FORMAT JSON)for understanding query performance - Connection testing that validates connectivity and reports latency
- Environment listing showing current host, database, and access-level configuration
- Optional HTTP bearer authentication, same-origin protection, request-size limits, and graceful shutdown
Installation
Download the latest release for your platform from the Releases page.
Usage
Environment Variables
| Variable | Required | Default | Description |
|---|---|---|---|
POSTGRESQL_HOST |
Yes | — | PostgreSQL hostname or IP |
POSTGRESQL_DATABASE |
Yes | — | Database name |
POSTGRESQL_USER |
Yes | — | Login username |
POSTGRESQL_PASSWORD |
Yes | — | Login password |
POSTGRESQL_PORT |
No | 5432 |
PostgreSQL port |
POSTGRESQL_SSLMODE |
No | prefer |
SSL mode: disable, allow, prefer, require, verify-ca, verify-full |
POSTGRESQL_ACCESS_LEVEL |
No | READONLY |
READONLY, DML-RW, or DDL-RW |
POSTGRESQL_CONNECTION_TIMEOUT |
No | 30 |
Connection timeout in seconds |
POSTGRESQL_QUERY_TIMEOUT |
No | 120 |
Query timeout in seconds |
POSTGRESQL_MAX_ROWS_DEFAULT |
No | 1000 |
Default row limit for queries |
POSTGRESQL_REQUIRE_CONFIRMATION |
No | true |
Require confirm flag for writes |
POSTGRESQL_TRANSPORT |
No | stdio |
MCP transport: stdio or http |
POSTGRESQL_HTTP_ADDR |
No | 127.0.0.1:8080 |
HTTP listen address when POSTGRESQL_TRANSPORT=http |
POSTGRESQL_HTTP_PATH |
No | /mcp |
Stateless Streamable HTTP endpoint path when POSTGRESQL_TRANSPORT=http |
POSTGRESQL_HTTP_BEARER_TOKEN |
No | — | Require Authorization: Bearer <token> for HTTP requests |
Running as an MCP Server
By default, the server communicates over stdio. Configure your MCP client to launch it:
{
"mcpServers": {
"postgresql": {
"command": "/path/to/bin/postgresql-mcp",
"env": {
"POSTGRESQL_HOST": "localhost",
"POSTGRESQL_DATABASE": "YourDatabase",
"POSTGRESQL_USER": "postgres",
"POSTGRESQL_PASSWORD": "YourPassword",
"POSTGRESQL_SSLMODE": "disable",
"POSTGRESQL_ACCESS_LEVEL": "READONLY"
}
}
}
}
To serve MCP using the stateless Streamable HTTP transport, set POSTGRESQL_TRANSPORT=http:
POSTGRESQL_TRANSPORT=http \
POSTGRESQL_HTTP_ADDR=127.0.0.1:8080 \
POSTGRESQL_HTTP_PATH=/mcp \
/path/to/bin/postgresql-mcp
Then configure a Streamable HTTP MCP client to connect to:
http://localhost:8080/mcp
Set POSTGRESQL_HTTP_BEARER_TOKEN when the endpoint is reachable by anything other than a trusted local client. Binding to a non-loopback address without authentication exposes every tool allowed by POSTGRESQL_ACCESS_LEVEL.
Access Levels
READONLY(default) — Schema exploration, data reading, profiling, relationship inspection, query explanation, connection testing, schema listing, extensions, views, triggers, CREATE TABLE DDL generation, table sizing. 18 tools.DML-RW— All read-only tools plusinsert_data,update_data,delete_data. 21 tools.DDL-RW— All DML tools pluscreate_table,create_index,drop_table. 24 tools.
Mutations (update_data, delete_data, drop_table) require a "confirm": true flag when POSTGRESQL_REQUIRE_CONFIRMATION is enabled (the default). Without confirmation, the server returns a preview of the affected rows instead.
update_data and delete_data predicates use named placeholders. For example, use "where": "id = $id AND tenant = $tenant" with "params": {"id": 42, "tenant": "acme"}. Every placeholder must have a value and every supplied parameter must be used.
Query results decode json/jsonb into structured JSON, encode bytea as base64, and suffix duplicate column names (id, id_2, and so on) so joined results do not silently lose values.
MCP Tools
Read-Only (READONLY)
| Tool | Description |
|---|---|
search_schema |
Search tables and columns by name pattern with pagination |
describe_table |
Get columns, primary keys, foreign keys, and indexes for a table |
list_table |
List tables, optionally filtered by schema and name |
list_databases |
List all databases on the server |
list_environments |
Show current connection and access-level configuration |
profile_table |
Row count, null counts, distinct counts, min/max when supported, with optional data samples |
inspect_relationships |
List foreign keys going out of and into a table |
inspect_dependencies |
Find objects (views, functions) that depend on a table |
explain_query |
Get the JSON execution plan for a read-only query |
read_data |
Execute a read-only SELECT query with row limits |
test_connection |
Ping the server and return latency and server version info |
validate_environment_config |
Validate that all environment variables are correctly configured |
list_schemas |
List all user schemas with owner and description |
list_extensions |
List installed PostgreSQL extensions with version |
list_views |
List views with their SQL definitions |
list_triggers |
List triggers with event, timing, and definition |
show_create_table |
Generate CREATE TABLE DDL including defaults, generated/identity columns, and table constraints |
table_size |
Table and index sizes, toast size, and estimated row counts, optionally for one table |
DML (DML-RW)
| Tool | Description |
|---|---|
insert_data |
Insert one or more rows into a table |
update_data |
Update rows matching a WHERE clause (with optional preview) |
delete_data |
Delete rows matching a WHERE clause (with optional preview) |
DDL (DDL-RW)
| Tool | Description |
|---|---|
create_table |
Create a table with column definitions, primary keys, and identity columns |
create_index |
Create a standard or unique index on specified columns |
drop_table |
Drop a table (with optional preview/confirmation) |
License
MIT License. See LICENSE for details.
PostgreSQL MCP questions
What tools does PostgreSQL MCP provide?
It exposes 24 tools, including search_schema, describe_table, list_table, list_databases, list_environments.
What access levels are available?
The server supports three tiered access levels: READONLY (default), DML-RW (adds insert/update/delete), and DDL-RW (adds create/drop table/index).
How can I ensure write operations are safe?
You can enable mutation confirmation with a preview mode that shows affected rows before executing writes by setting POSTGRESQL_REQUIRE_CONFIRMATION to true.
Can I use this server over HTTP?
Yes, by setting the POSTGRESQL_TRANSPORT environment variable to 'http', you can serve the MCP via a stateless Streamable HTTP endpoint.
How does the server handle large query results?
The server uses authoritative row limits to cap results from the read_data tool, even if the query itself contains a larger LIMIT clause.