GibsonAI MCP Server

What is GibsonAI MCP Server about?

GibsonAI MCP Server enables developers to manage GibsonAI projects and hosted databases through natural‑language commands. It acts as a bridge between MCP‑compatible clients (Cursor, Windsurf, Claude Desktop, VS Code, Cline) and the GibsonAI platform, allowing creation, inspection, modification, and deployment of relational schemas without writing boilerplate code.

How to use GibsonAI MCP Server?

Authenticate with Gibson CLI – run uvx --from gibson-cli@latest gibson auth login to store your API credentials.
Start the MCP server – the server is launched by the CLI command uvx --from gibson-cli@latest gibson mcp run.
Configure a client – add a new MCP server entry in the client’s configuration (Cursor, Windsurf, Claude Desktop, VS Code, or Cline) using the same command and arguments shown above.
Issue natural‑language prompts – ask the client to create schemas, run queries, deploy projects, etc.; the server translates the request into GibsonAI API calls and returns results.

Key Features of GibsonAI MCP Server

Project Management – get_projects, create_project, update_project, get_project_details.
Schema Operations – get_project_schema, get_deployed_schema, submit_data_modeling_request for creating and modifying models via plain language.
Database Interaction – query_database executes SQL against hosted databases; get_project_hosted_database_details supplies connection info.
Deployment Automation – deploy_project triggers migrations and pushes changes to development or production environments.
Cross‑IDE Support – ready‑to‑use configurations for Cursor, Windsurf, Claude Desktop, VS Code (Copilot), and the Cline VS Code extension.

Use Cases

Rapid prototyping – describe a new application schema (“blogging platform with users, posts, comments”) and get a fully formed relational model instantly.
Schema evolution – add foreign keys, rename columns, or create indexes by describing the change in natural language; the server generates and applies migration scripts.
Data exploration – run ad‑hoc SQL queries against the hosted database without leaving the editor.
Continuous deployment – push schema updates to staging or production with a single command (deploy_project).
Mock data generation – seed tables with realistic data for testing using natural‑language prompts.

FAQ

Q: Do I need to install anything besides the Gibson CLI? A: No. The MCP server is bundled with the CLI; once authenticated, the uvx … gibson mcp run command starts the server.

Q: Which clients are supported? A: Cursor, Windsurf, Claude Desktop, VS Code (Copilot), and the Cline VS Code extension all support the required MCP configuration.

Q: How are credentials handled? A: Authentication is performed once with gibson auth login. The CLI stores a token that the MCP server uses for all subsequent API calls.

Q: Can I run the server locally for development? A: Yes. The server runs as a standard STDIO process started by the uvx command, making it easy to test locally.

Q: What if I need to change the project name? A: Use the update_project tool with the project's UUID; only the project_name field can be modified.

The GibsonAI Model Context Protocol Server provides a set of tools to MCP Clients like Cursor, Windsurf, or Claude Desktop. These clients can use these tools to interact with your GibsonAI projects and databases using your natural language instructions.

You can accomplish various tasks with GibsonAI directly in your favorite IDE, for example:

Create new GibsonAI projects and design database schemas
View project structure, schema diagrams, a summary of tables and relationships
Apply schema changes and trigger automatic migrations
Run SQL queries against your database
Deploy projects to development or production environments
Seed tables with mock data
Build a full-stack apps

Prompt Examples:

“Create a blogging platform schema with users, posts, and comments.”
“Add a foreign key from bookings to payments.”
“Generate mock data for the boooking destination table.”
“Fetch connection string for my blogging database”
“Explain how the tables are related in this project.”

Usage Examples

Authentication

You'll need to ensure you're logged in to the Gibson CLI before the MCP server will work.

uvx --from gibson-cli@latest gibson auth login

Cursor Setup

Click the Add to Cursor button above or go to Cursor → Settings → Cursor Settings → MCP Tools and click New MCP Server. Update the configuration to include the following:

{
  "mcpServers": {
    "gibson": {
      "command": "uvx",
      "args": ["--from", "gibson-cli@latest", "gibson", "mcp", "run"]
    }
  }
}

Windsurf Setup

Go to Windsurf → Settings → Windsurf Settings → Cascade and click Add server in the Model Context Protocol (MCP) Servers section

In the modal, click Add custom server

Update the configuration to include the following:

{
  "mcpServers": {
    "gibson": {
      "command": "uvx",
      "args": ["--from", "gibson-cli@latest", "gibson", "mcp", "run"]
    }
  }
}

Open the Cascade chat and, if necessary, refresh the MCP servers

Claude Desktop Setup

Go to Claude → Settings → Developer and click Edit Config

Open the claude_desktop_config.json file and update the configuration to include the following:

{
  "mcpServers": {
    "gibson": {
      "command": "uvx",
      "args": ["--from", "gibson-cli@latest", "gibson", "mcp", "run"]
    }
  }
}

See the Claude Desktop MCP docs for more information.

Claude Code Setup

claude mcp add gibson -- uvx --from gibson-cli@latest gibson mcp run

claude mcp get gibson

gibson:
  Scope: Local (private to you in this project)
  Type: stdio
  Command: uvx
  Args: --from gibson-cli@latest gibson mcp run
  Environment:

To remove this server, run: claude mcp remove "gibson" -s local

VS Code + GitHub Copilot Setup

Create or open the .vscode/mcp.json file

Update the configuration to include the following:

{
  "inputs": [],
  "servers": {
    "gibson": {
      "type": "stdio",
      "command": "uvx",
      "args": ["--from", "gibson-cli@latest", "gibson", "mcp", "run"]
    }
  }
}

See the official GitHub Copilot MCP docs for more information.

Cline (VS Code Extension) Setup

Open Cline in VS Code:
Go to Sidebar → Cline icon.
To configure MCP Servers in Cline, you need to modify the cline_mcp_settings.json file. Click the MCP Servers icon → go to Installed → click Configure MCP Servers to open the configuration file.
Add the following gibson server entry inside the mcpServers object:

{
  "mcpServers": {
    "gibson": {
      "command": "uvx",
      "args": ["--from", "gibson-cli@latest", "gibson", "mcp", "run"]
    }
  }
}

Save the file. Cline should reload the configuration automatically.

🔧 Supported Tools

🗂 `get_projects`

Title: List all existing projects
Description: Retrieves all GibsonAI projects associated with the authenticated user. Useful when the user refers to a project by name but you need the UUID. If a .gibsonai file exists, use it instead unless the user intends otherwise.

🆕 `create_project`

Title: Create a new project
Description: Creates a new GibsonAI project. Check for an existing .gibsonai file or similar project names before creation. Prompt the user to update or create the .gibsonai file with the new UUID.

🔍 `get_project_details`

Title: Fetch project metadata
Description: Returns metadata and configuration for a given project using its UUID. Ideal when working with an existing .gibsonai file to load project-specific context.

🔗 `get_project_hosted_database_details`

Title: Get hosted database connection details
Description: Returns credentials, connection string, dialect, and other necessary details for querying the hosted GibsonAI database. Useful for building queries or integrating with tools.

✏️ `update_project`

Title: Rename a project
Description: Updates the project name using its UUID. Currently, only the project_name field is supported.

🧠 `submit_data_modeling_request`

Title: Submit schema modeling request
Description: Submit any natural-language data modeling request (e.g., create, modify schema). This tool fully handles the request using GibsonAI's internal modeler and should be used instead of any manual schema design.

🚀 `deploy_project`

Title: Deploy to database(s)
Description: Triggers automatic schema migrations and deploys the current schema to all GibsonAI supported databases.

📐 `get_project_schema`

Title: Get working schema
Description: Retrieves the current state of the schema including unpublished or un-deployed changes.

✅ `get_deployed_schema`

Title: Get live schema
Description: Fetches the schema currently deployed to the primary hosted database. Use this to compare against the working schema or confirm deployment to your primary database (e.g. Production)

🧾 `query_database`

Title: Run SQL queries
Description: Runs the provided SQL query against a database by using the API key associated with that database. Ensure correct quoting for identifiers depending on the SQL dialect (e.g., backticks for MySQL, double quotes for PostgreSQL).

Distribution

Note that this repo is for documentation purposes only. Our MCP server code lives within our CLI, which allows us to share authentication + API interaction logic with the CLI and have a single distribution. This means we're able to ship new features to you faster.

GibsonAI MCP Server