Antigravity GLM

You can run the Antigravity GLM MCP server to bridge Gemini with the GLM-4.5 model, enabling automated, tool-driven workflows with strong security and reliability features. This MCP server lets you delegate complex tasks to GLM, manage memory and backups, and limit risky shell commands while exposing a powerful set of 25 tools for seamless automation.

How to use

Set up the MCP server to run locally or in your environment, then connect your MCP client to the local server using the standard MCP protocol configured for stdio in your environment. Start the server from your virtual environment and point your client to localhost or the appropriate host:port if you expose the MCP server over the network. The server exposes 25 tools that you can invoke from your MCP client to delegate tasks to GLM-4.5, perform file operations, run code in a sandbox, query memory, and manage data with automatic backups and secure execution.

How to install

Prerequisites: ensure you have Python 3.11 or newer installed. Create and activate a virtual environment for isolation.

Steps to install and run the MCP server locally are provided below. Follow these steps in order to get the server up and running quickly.

Install steps

# 1. Clone the project repository
git clone https://github.com/coreline-ai/antigravity_glm_mcp.git
cd antigravity_glm_mcp

# 2. Create and activate a virtual environment
python3.11 -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate

# 3. Automated installation (recommended)
python scripts/install.py

Manual installation (alternative)

# Install dependencies
pip install -r requirements.txt

# MCP configuration file placement (example path)
# Add the following to your MCP settings file (e.g., ~/.gemini/settings.json)

The following example shows how to configure the MCP entry directly in a settings file. This is tailored for a local stdio-based run using a Python interpreter from a virtual environment and the main server script.

{
  "mcpServers": {
    "antigravity_glm_mcp": {
      "command": "/path/to/.venv/bin/python",
      "args": ["/path/to/antigravity_glm_mcp/src/server.py"],
      "env": {
        "PROJECT_ROOT": "/your/workspace",
        "ZHIPU_API_KEY": "your-api-key",
        "GLM_MODEL": "GLM-4.5",
        "GLM_BASE_URL": "https://api.z.ai/api/coding/paas/v4",
        "PYTHONPATH": "/path/to/antigravity_glm_mcp"
      }
    }
  }
}

Security and usage notes

The server enforces a multi-layer security approach. Only commands registered in the whitelist can be executed, and sensitive environment variables are isolated to prevent leakage. Backups are automatically created before modifications, and a permanent memory store preserves context across sessions.

Environment and troubleshooting

Ensure your environment variables are set as shown in the example configuration. If you encounter startup issues, verify the virtual environment is active, and the Python path in the command matches your setup. Check that the API key and model settings are correct, and that the base URL points to the GLM service you intend to use.

Available tools

glm_cmd

Delegates complex tasks to GLM with context and task instructions. Enables structured prompting and controlled delegation.

glm_bypass

Sends raw prompts directly to GLM, bypassing higher-level abstractions for specialized use cases.

glm_image_analyze

Performs image analysis using Vision capabilities with image_path and prompt parameters.

glm_file_read

Reads file contents with path and encoding options.

glm_file_create

Creates a new file with specified content and overwrite behavior.

glm_file_edit

Edits a file by replacing old_string with new_string.

glm_file_delete

Deletes a file and preserves a backup.

glm_file_rollback

Restores a previous version of a file by path and version.

glm_dir_list

Lists directory contents with path and optional recursive traversal.

glm_grep

Searches files using a regular expression pattern.

glm_code_run

Executes Python code in a sandbox with a timeout.

glm_shell_exec

Runs whitelisted shell commands from a controlled whitelist.

glm_git_status

Shows the status of a Git repository at a given path.

glm_git_commit

Commits changes with a message and optional add_all flag.

glm_git_log

Displays Git commit history with options for count and formatting.

glm_git_diff

Shows differences between working tree states with options.

glm_http_request

Performs HTTP requests with SSRF protection and method/body controls.

glm_web_search

Executes web searches via DuckDuckGo with query and result limits.

glm_memory_save

Saves persistent memory under a key with category tagging.

glm_memory_get

Retrieves a memory value by key.

glm_memory_list

Lists memory entries by category with limit control.

glm_memory_delete

Deletes a memory entry by key.

glm_db_query

Runs SQLite queries against a specified database path.

glm_schedule_task

Schedules a task with cron-like expressions and actions.

glm_action_log

Provides logs of agent actions with filters for limit and type.