home / mcp / gibsonai mcp server

GibsonAI MCP Server

GibsonAI's MCP server

Installation

Add the following to your MCP client configuration file.

Configuration

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

The GibsonAI MCP Server gives you a programmable bridge to interact with GibsonAI projects and their databases using natural language prompts from your MCP clients. It powers tools that let you create, view, modify, and deploy your data models, run queries, and seed data directly from your preferred IDE or chat interface.

How to use

Connect your MCP clients to the GibsonAI MCP Server to start managing projects and schemas through natural language prompts. You can use clients like Cursor, Windsurf, Claude Desktop, VS Code with GitHub Copilot, or Cline to interact with GibsonAI projects and databases. Typical workflows include creating new projects, viewing project structure, applying schema changes with migrations, running SQL queries, and deploying schemas to your databases. Use the provided MCP server entry points in each client to establish the connection, then issue commands or prompts to manage your models and data.

How to install

Prerequisites: you need access to the Gibson CLI tooling and authentication setup prior to starting the MCP server. Ensure you are logged in to the Gibson CLI before the MCP server will work.

Step 1. Log in to the Gibson CLI to enable MCP interactions.

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

Step 2. Start the MCP server using the standard runtime command shown in client configurations. The server runs as a local stdio process connected to your GibsonAI environment.

Configuration and usage notes

You can connect MCP clients to the GibsonAI MCP Server using the following stdio configuration snippet. This config runs the MCP server in-process via the uvx runtime and points to the Gibson CLI for authentication and command routing.

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

Troubleshooting and tips

If you encounter issues connecting, verify that you are authenticated with the Gibson CLI and that the uvx runtime command is accessible in your shell. Ensure the MCP server configuration is loaded in your MCP client (Cursor, Windsurf, Claude Desktop, or Cline) and that the client has permissions to access GibsonAI projects.

Notes and examples

The GibsonAI MCP Server exposes a set of tools that enable you to list, create, query, and deploy GibsonAI projects and their schemas. You can also retrieve project metadata, host database connection details, and run SQL against your databases through these endpoints.

Available tools

get_projects

List all existing GibsonAI projects associated with the authenticated user and provide UUIDs for reference.

create_project

Create a new GibsonAI project and prompt updates to the local project file as needed.

get_project_details

Fetch metadata and configuration for a given project by its UUID.

get_project_hosted_database_details

Return credentials, connection string, dialect, and other details for querying the hosted GibsonAI database.

update_project

Rename a project by its UUID, currently supporting updating the project_name field.

submit_data_modeling_request

Submit natural-language data modeling requests; this tool handles schema changes end-to-end.

deploy_project

Trigger automatic migrations and deploy the current schema to all supported databases.

get_project_schema

Get the working schema, including unpublished or un-deployed changes.

get_deployed_schema

Get the live deployed schema on the primary hosted database for comparison.

query_database

Run SQL queries against a database using the associated API key and proper SQL quoting.