Note
With the release of MagicPod version 1.69.0, the MCP server has been redesigned. As a result, existing users of the previous MCP server must update their configuration. Please follow the setup instructions below to migrate to the new MCP server.
Table of contents
Overview
MagicPod MCP Server is a module that enables AI agents (Cline, Cursor, Claude, etc.) to operate various MagicPod functions. By implementing this, you can automate MagicPod operations by giving instructions to AI agents in natural language.
Some of the main operations currently available through the MagicPod MCP server include:
- Creating, editing, viewing, and running test cases
- Creating, editing, and viewing shared steps
- Retrieving test run results
- Retrieving UI information and validating locators
- Searching MagicPod Help Center articles
When creating or editing test cases, you can also have the AI read files such as Excel spreadsheets and give it instructions based on their contents. This enables workflows such as:
- Have the AI read test procedures created in Excel or a similar tool and create the necessary test cases.
- Have the AI read specifications for a new feature and update the affected test cases.
- Analyze the cause of a test failure and update the test case as necessary.
For specific examples of what you can do with the MagicPod MCP server, see also Prompt Examples.
The full list of available operations (tools) is provided below.
| Tool | Description |
capture_ui |
Capture the current UI of the device |
create_shared_step |
Create a shared step from a consecutive range of steps in the current test case |
create_testcase |
Create a new empty test case in the specified project |
edit_data_pattern |
Edit the data patterns (values used for data-driven testing) of the current test case |
edit_shared_step_parameter |
Add or remove parameters for the current shared step |
edit_shared_step_steps |
Edit test steps in a shared step |
edit_steps |
Edit test steps in the current test case |
edit_ui_elements |
Edit the properties (name and locator) of UI elements on a saved UI |
execute_partial_steps |
Run part of a test by specifying the start and end steps |
find_help_center_article |
Search and view MagicPod Help Center articles |
get_command_definition |
Retrieve detailed definitions of one or more MagicPod commands |
get_health_score |
Retrieve the health score of a project |
get_shared_step |
Retrieve details of a single shared step, including its steps in a human-readable format |
get_shared_step_internal_steps |
Retrieve test steps in a shared step definition |
get_test_case_detail |
Retrieve details of a single test case, including its steps in a human-readable format |
get_test_run |
Retrieve information and artifacts for a single test run |
inspect_locators |
Suggest stable locators for a screen or validate existing locators |
inspect_ui |
Retrieve and search previously captured UIs (screens) for the device in the current session |
list_batch_run_test_runs |
List all test runs in a batch run, grouped by test pattern |
list_organizations |
List organizations you can access |
list_projects |
List projects you can access within an organization |
list_test_runs |
List test runs for a test case in descending run number order (newest first) |
magicpod_list |
List test cases, shared steps, batch runs, and test settings in a project |
manage_device |
Start or stop a device for the specified session |
read_shared_step |
Load the steps and parameter definitions of the current shared step |
read_test_case |
Load data for the current test case |
search_shared_steps |
Search shared steps by name or description using a keyword or regular expression |
start_batch_run |
Start a batch run based on saved test settings |
start_session |
Start a MagicPod session for the specified test case |
start_shared_step_session |
Start a MagicPod session for editing the specified shared step |
trash_shared_step |
Move a shared step to the Trash |
Future expansion plans: We plan to expand functionality in the future, including enhanced automatic repair. Reference: Press Release(Japanese)
Setup Instructions
Support for connections using OAuth authentication is coming soon.
Prerequisites
To use the MagicPod MCP Server, you need:
- A MagicPod account with API access enabled
- A MagicPod API token
-
Node.js (only for AI clients that use
npx)
Since the MagicPod MCP Server runs on MagicPod's infrastructure, your AI client connects directly to it. No additional software installation is required.
Connect Your AI Client
Claude Desktop
- Open Settings > Developer > Edit Config.
-
Add the following configuration under
"mcpServers". If"mcpServers"does not already exist, create it.{ "mcpServers": { "magicpod": { "command": "npx", "args": [ "mcp-remote", "https://app.magicpod.com/mcp", "--header", "Authorization:Bearer YOUR_API_TOKEN" ] } } } - Save the file and restart Claude Desktop.
Claude Code
Run the following command in your terminal:
claude mcp add --transport http magicpod https://app.magicpod.com/mcp --header "Authorization: Bearer YOUR_API_TOKEN"
Copilot CLI
Run the following command in your terminal:
copilot mcp add --transport http magicpod https://app.magicpod.com/mcp --header "Authorization: Bearer YOUR_API_TOKEN"
Codex
-
First, set your MagicPod API token as an environment variable.
export MAGICPOD_API_TOKEN=YOUR_API_TOKEN
-
Then run the following command in your terminal:
codex mcp add magicpod --url https://app.magicpod.com/mcp --bearer-token-env-var MAGICPOD_API_TOKEN
To keep the environment variable across terminal sessions, add the export command to your shell profile (for example, ~/.zshrc or ~/.bashrc).
Cursor
- Open the Cursor MCP configuration file.
- For all projects:
~/.cursor/mcp.json - For a specific project only:
<project-root>/.cursor/mcp.json
- For all projects:
-
Add the following configuration under
"mcpServers":{ "mcpServers": { "magicpod": { "command": "npx", "args": [ "mcp-remote", "https://app.magicpod.com/mcp", "--header", "Authorization:Bearer YOUR_API_TOKEN" ] } } } - Restart Cursor.
Gemini — Antigravity IDE / Antigravity 2.0
Note
The browser version of Gemini and Gemini CLI do not support remote MCP servers. To use the MagicPod MCP Server with Gemini, use Antigravity IDE or Antigravity 2.0.
- Open: Settings > Customizations > Open MCP Config
-
Add the following configuration:
{ "mcpServers": { "magicpod": { "command": "npx", "args": [ "mcp-remote", "https://app.magicpod.com/mcp", "--header", "Authorization:Bearer YOUR_API_TOKEN" ] } } } - Save the configuration and restart the IDE.
ChatGPT
Note
The browser version of ChatGPT does not currently support the MagicPod MCP Server. Support is planned in a future update.
Configure Tool Permissions
Depending on the AI agent you use, you may be able to configure permissions for each tool provided by the MagicPod MCP server.
Restricting certain tools may limit some operations or reduce accuracy. Therefore, we generally recommend enabling all tools and setting them to Always allow.
However, your organization may have security policies or operational rules regarding the use of AI agents. Configure tool permissions as necessary in accordance with your organization's policies and rules.
Troubleshooting
-
Connection or Authentication Errors
Verify that your API token is correct, your API token has not expired, or your network can access
https://app.magicpod.com.If necessary, generate a new API token from the MagicPod settings page.
-
Displayed
npx: command not foundInstall Node.js 18 or later.
npxis included with Node.js. -
Error Displayed During Response Generation
For example, when using the free version of Claude, you may see an error like "Claude's response was interrupted" during response generation. This is due to character limits in the free version and can be avoided by using the paid Claude Pro version. Similar issues may occur with other AI agents, so consider subscribing to paid plans.
Security
MagicPod MCP Server is currently a very simple module that connects AI agents with MagicPod services (MagicPod Web API and Help Center). The repository is publicly available on GitHub.
- Your information will not be used for machine learning.
- While the Web API token needs to be written in the MCP server's config file, it is designed not to pass this information to AI agents.
- We plan to expand functionality in the future while maintaining a security-conscious configuration.