Antigravity Plugin for Harness
Install the Antigravity Plugin for Harness to connect your AI agent to Harness and use workflow skills for pipelines, deployments, and security
The Antigravity Plugin for Harness connects Google Antigravity to your Harness account through the hosted Model Context Protocol (MCP) server. Use natural language to investigate pipeline failures, review deployments, and draft configuration with 71 bundled Harness workflow skills.
Go to the Antigravity Plugin for Harness repository to download the plugin and review its source.
What you will learn from this topic
By the end of the topic, you will be able to:
Install the plugin from the repository using the Antigravity CLI.
Authenticate with Harness through OAuth.
Verify the connection with a read-only request.
Use Harness workflow skills in your agent conversations.
Before you begin
Before you istall the Antigravity Plugin for Harness, ensure you have the following:
Antigravity CLI (
agy). Go to Google's CLI installation guide to install it for your operating system.Git to clone the plugin repository.
Harness account with access to the organizations, projects, and resources you want to use.
Hosted MCP OAuth enabled for your Harness account. Go to Harness Support to request enablement if it is not already available.
The default connection uses hosted OAuth at https://mcp.harness.io/mcp. You do not need a personal access token (PAT), a local MCP server, or Python for this connection.
Install and connect
Step 1: Install the plugin
Clone the repository:
Validate the package:
Install the plugin locally:
Verify the installed state:
Confirm that harness-antigravity is enabled and that its inventory includes 71 skills, one rule, and the harness MCP server. Go to Client compatibility to review known validation warnings.
Step 3: Authenticate with Harness
Enter /mcp in the interactive CLI. Under Plugins, select the Harness server. In CLI 1.4.2, its name is harness-antigravity_harness.
Select Authenticate and complete Harness sign-in in the browser. If the callback displays an authorization code, paste it into the CLI's authorization-code field.
Return to the MCP panel. Confirm that the server shows Authed and that Harness tools are listed.
Keep authorization codes and tokens out of agent conversations, repository files, and shared logs.
Step 4: Verify a read-only request
Ask the agent to list projects:
A successful response containing Harness project data verifies API access. An installed plugin or an Authed label alone does not verify access to a particular resource.
For subsequent requests, specify the account, organization, and project you intend to use.
Use Harness in Antigravity
Describe the outcome you want in the agent conversation. Bundled skills provide Harness-specific procedures, while MCP tools read or act on Harness resources using your authenticated permissions. Go to the skill catalog to explore the workflows and their prerequisites.
Pipeline troubleshooting
In account <account>, organization <org>, project <project>, inspect the latest failed execution of pipeline <pipeline> and explain the failure. Do not rerun it.
Pipeline authoring
Draft a CI pipeline for this Node.js repository. Show me the YAML and required connectors before saving anything to Harness.
Deployment readiness
Review deployment readiness for service <service> in project <project> and organization <org>. Report missing dependencies without changing resources.
Security investigation
Summarize security findings for project <project> in organization <org>. Group them by severity and recommend next steps. Do not change exemptions.
Feature management
Review the targeting configuration for flag <flag> in the specified Harness scope and explain its behavior without changing it.
Available operations depend on your Harness modules, resource permissions, and each skill's prerequisites. Review proposed changes and the target scope before approving write or execution requests.
Client compatibility
Installation, discovery of all 71 skills, hosted OAuth, and a read-only project listing were verified with Antigravity CLI 1.4.2. This smoke test does not establish that every bundled workflow has been tested end to end.
Native validation passes with warnings in this client. Strict validation reports five warnings: the two oversized skills, the root development AGENTS.md, and the manifest fields category and mcpServers. The runtime rule in rules/AGENTS.md and the default mcp_config.json are discovered successfully.
Go to the repository's validation status to review recorded checks and remaining limitations.
Troubleshooting
Next steps
Plugin repository: Review updates and connection options, including the optional local PAT profile.
Harness MCP Server: Explore the server and supported configuration options.
Harness Skills: Understand how workflow skills guide agent tasks.
Last updated
Was this helpful?