Use Gfs MCP

GFS MCP Server

GFS provides an MCP (Model Context Protocol) server for AI agent integration, enabling programmatic access to database version control operations including commits, branches, time travel, and schema versioning.

Supported Databases

• PostgreSQL (versions 13-18)
• MySQL (versions 8.0-8.1)
• SQLite (version 3) — file-based; needs no container runtime

With SQLite, init, commit, log, status, checkout, extract_schema, show_schema, diff_schema, export_database, import_database and query all work with no container runtime running. compute and user do not apply: there is no server to start and no roles to manage.

Installation

1. Verify GFS CLI is installed

The MCP server requires the GFS CLI to be installed on the host system.

2. Configure MCP Server

Add GFS to your MCP client configuration (e.g., Claude Desktop, custom MCP client). The MCP server runs as: gfs mcp

Configuration requires:

• Command: Path to gfs binary
• Args: ["mcp"]
• Optional: GFS_REPO_PATH environment variable

Quick Start

1. Start MCP Server

The server starts automatically when configured in your MCP client. It provides access to all GFS operations through MCP tools.

2. Check Available Tools

The MCP server exposes 14 tools for database version control operations. All tools accept an optional path parameter to specify repository location.

3. Verify Repository Status

Use the status tool to check repository state, current branch, HEAD commit, and database container status.

Revision References

GFS supports Git-style revision notation in all tools that accept commit references (checkout, show_schema, diff_schema, log):

Basic Formats

• HEAD - Current commit
• main - Branch tip (any branch name)
• abc123def456... - Full commit hash (64 characters)

Tilde Notation (Ancestor References)

Reference ancestor commits using ~ notation:

• HEAD~1 - Parent of HEAD (previous commit)
• HEAD~5 - 5th ancestor of HEAD
• main~3 - 3 commits before main branch tip
• abc123~2 - 2nd parent of commit abc123
• HEAD~ - Same as HEAD~1 (defaults to 1)

MCP Tool Usage Examples

// Checkout previous commit
checkout({ revision: "HEAD~1" })

// Compare schema with 5 commits ago
diff_schema({ commit1: "HEAD~5", commit2: "HEAD" })

// View schema from 3 commits back
show_schema({ commit: "main~3" })

// Show ancestor schema
show_schema({ commit: "abc123def456...~2" })

// View log from ancestor to current
log({ from: "HEAD~10", until: "HEAD" })

Core Workflow

Best Practices

CRITICAL: Always use the commit tool before database mutations. This enables rollback if issues occur.

Commits automatically capture database schema at commit time, providing complete versioning of both data and structure.

1. List Providers

Tool: list_providers

Lists all supported database providers, versions, and available features. No parameters required.

Returns: Array of providers with supported versions and feature lists.

2. Initialize Repository

Tool: init

Creates a new GFS repository with specified database provider and version.

Parameters:

• database_provider (required) - Provider name (postgres, mysql)
• database_version (required) - Version string (e.g., "17", "8.0")
• path (optional) - Repository location

Returns: Repository path and initialization status.

3. Check Status

Tool: status

Gets repository and database container status including current branch, HEAD commit, and connection information.

Parameters:

• path (optional) - Repository location

Returns: current_branch, head_commit (absent before the first commit), active_workspace_data_dir, and compute when a container is configured. For SQLite, connection_string is a sqlite: URL to the file in the active workspace and there is no compute section.

gfs status reports the same fields from the same source, so the CLI and this tool cannot disagree.

4. Commit Changes

Tool: commit

Creates snapshot of database state, automatically capturing schema at commit time.

Parameters:

• message (required) - Commit message
• path (optional) - Repository location
• author (optional) - Author name
• author_email (optional) - Author email

Returns: Commit hash.

Automatic Schema Versioning: Every commit captures database schema including structured metadata and native DDL.

5. View History

Tool: log

Views commit history with optional filtering.

Parameters:

• path (optional) - Repository location
• max_count (optional) - Limit number of commits
• from (optional) - Starting revision
• until (optional) - Ending revision

Returns: Array of commits with metadata, timestamps, authors, and references.

6. Create and Switch Branches

Tool: checkout

Switches to existing branch, specific commit, or creates new branch.

Parameters:

• revision (required) - Branch name or commit hash
• path (optional) - Repository location
• create_branch (optional) - Name of a new branch to create and switch to. A string, not a boolean: create_branch: true is rejected with invalid type: boolean, expected a string

Returns: Success status and checked out revision.

Tip: For branch listing and creation without switching, use the CLI with --json:

# List branches programmatically
gfs branch

# Create branch from checkpoint without switching
gfs branch release/v1.0 <commit_hash>

# Delete a branch
gfs branch -d old-feature

7. Database Container Management

Tool: compute

Manages database container lifecycle.

Parameters:

• action (required) - Operation: status, start, stop, restart, pause, unpause, logs
• path (optional) - Repository location
• logs_tail (optional, logs only) - Number of lines to show
• logs_since (optional, logs only) - Timestamp filter

Returns: Status information or log output based on action.

Actions:

• status - Container state and metadata
• start - Start stopped container
• stop - Stop running container
• restart - Restart container
• pause - Pause container (keeps state in memory)
• unpause - Resume paused container
• logs - View container logs

8. Query Database

Tool: query

Executes SQL queries against the database or returns connection information.

Parameters:

• query (optional) - SQL query to execute (omit for connection info)
• path (optional) - Repository location
• database (optional) - Override database name. Not accepted for SQLite: the file is the database, so there is nothing to select, and passing it is an error rather than a silently ignored argument.

Returns: Query results or connection information if no query provided. For SQLite the connection information is a sqlite: URL to the file in the active workspace, with no host or port.

This tool runs the database's own client on the host — psql, mysql, sqlite3 — so that binary must be installed. Nothing else in this list needs it: GFS reads SQLite through a linked engine.

9. Schema Operations

Extract Current Schema

Tool: extract_schema

Extracts schema from running database, returning structured metadata.

Parameters:

• path (optional) - Repository location

Returns: Complete schema metadata including schemas, tables, columns, constraints, and relationships as structured JSON.

Use case: Understand current database structure before making changes or writing queries.

Show Schema from Commit

Tool: show_schema

Views database schema as it existed at any commit in history.

Parameters:

• commit (required) - Commit hash or reference (HEAD, main, branch name)
• path (optional) - Repository location
• metadata_only (optional) - Return only JSON metadata
• ddl_only (optional) - Return only SQL DDL

Returns: Schema object with hash, metadata, and/or DDL based on flags.

Output includes:

• Schema hash (content-addressed identifier)
• Database driver and version
• Structured metadata (schemas, tables, columns with types and constraints)
• Native DDL dump (CREATE TABLE statements)

Use case: View schema at any point in history, document database structure, audit schema changes.

Compare Schemas

Tool: diff_schema

Tracks schema evolution by comparing schemas between two commits.

Parameters:

• commit1 (required) - First commit hash or reference
• commit2 (required) - Second commit hash or reference
• path (optional) - Repository location

Both commit1 and commit2 accept rev~n notation for ancestor references (e.g., HEAD~5, main~3).

Returns: Schema comparison with hashes, table/column count changes, and difference summary.

Comparison includes:

• Schema hash equality check
• Table count changes
• Column count changes
• Summary of additions/removals

Use case: Review schema changes before merging branches, track schema evolution, identify breaking changes.

10. Import/Export Data

Export Database

Tool: export_database

Exports database data to file.

Parameters:

• format (required) - Export format (sql, custom)
• path (optional) - Repository location
• output_dir (optional) - Output directory path

Returns: Export file path and status.

Supported formats:

• sql - Plain SQL dump
• custom - Database-specific binary format (PostgreSQL pg_dump custom format)

Import Database

Tool: import_database

Imports data from file into database.

Parameters:

• file (required) - Path to import file
• path (optional) - Repository location
• format (optional) - Data format (sql, csv, json, custom)

Returns: Import status.

Supported formats: SQL, CSV, JSON, custom (database-specific formats).