Masthead MCP
Masthead gives your agent live incidents, lineage and cost data from your BigQuery estate, so Claude Code, Codex or Antigravity can triage and fix in minutes.
Works in Claude Code, OpenAI Codex and Google Antigravity.
/plugin marketplace add masthead-data/for-agents/plugin install masthead-agent-tools@masthead-dataRestart Claude Code, run /mcp and authenticate masthead.
codex plugin marketplace add masthead-data/for-agentscodex plugin add masthead-agent-tools@masthead-datacodex mcp login mastheadRestart Codex after installing the plugin.
git clone https://github.com/masthead-data/for-agents.git ~/.gemini/config/plugins/masthead-agent-toolsOr clone it into .agents/plugins/masthead-agent-tools inside your project.
Restart Antigravity, then open Agent Settings, Customizations, masthead and choose Authenticate.
Prefer a shortcut? The Agent tools button in the Masthead app copies a setup prompt that walks your agent through these steps. Ask to verify Masthead MCP connection after the setup.
Read the docsWhen something breaks, the answer is spread across your warehouse console, your orchestrator, your BI tool and a Slack thread. By the time you have it, stakeholders are already asking.
The root cause hides across job logs, pipeline code and reports, and every check is another tab.
You find out which reports broke from the people who read them.
Unused tables and idle pipelines keep billing, because finding them is a project of its own.
Masthead already has the answers. Now you can ask for them where you work.
Your agent reads the incident, traces the affected tables and proposes a fix for you to review.
Illustrative example
Masthead gives your agent the context. Your agent proposes the change, and you decide when to run it.
Install the pluginmodels/staging/stg_orders.sql
select order_id, customer_id, order_total, total_amount as order_total, created_atfrom {{ source('raw_shop', 'orders') }}
dbt run passed after the fixIncident resolved
The Masthead plugin gives your agent real incidents, lineage, ownership and cost data, so it stops guessing and starts working.
Your agent spots dead-end and unused tables, wasteful pipelines, tables on the wrong storage billing model and mis-sized slot commitments, then hands you a review list with dry-run commands. You run them, not the agent.
/masthead-agent-tools:savingsYour agent pulls the anomaly, its lineage and real downstream usage, ranks it by business impact, drafts a fix for review and routes the rest to the owning team.
/masthead-agent-tools:triageYour agent ranks tables and BI assets by their asset score, suggests PRIORITY or CRITICAL tiers with reasons, and applies them once you confirm.
Describe the check in plain words. Your agent builds the metric table, schedules its refresh, gives Masthead read access and creates the scan, once you confirm. Later it can list, pause, update or delete your scans.
Your agent investigates and recommends. Every cost-saving workflow stops before any destructive command and leaves the final call to you.
The masthead-agent-tools plugin bundles the skills and the Masthead MCP server. These two workflows run end to end, so you can start with one of them.
Open incidents, lineage, blast radius and a mitigation plan. Takes an optional scope, such as a project name or an incident UUID.
/masthead-agent-tools:triageDead-end and unused tables, billing models, pipelines, workload assignments and slot commitments. Reads the insights table from the BigQuery export.
/masthead-agent-tools:savingsEverything you need to know before you add Masthead to your agent. For anything not covered here, check the docs or contact us directly.
Claude Code, OpenAI Codex and Google Antigravity, through the masthead-agent-tools plugin.
Yes. The plugin works with your Masthead tenant, and you sign in once. The FinOps workflows also need the BigQuery insights export enabled. If it is off, the skills link you to a request-access page.
The cost skills only investigate and recommend. They prepare dry-run commands for you to review and never run destructive ones. Some MCP tools can update objects inside Masthead, such as an incident’s status or assignee, asset tiers and data scans, and only after you confirm.
Masthead’s observability works from BigQuery logs and metadata. It doesn’t read or edit your data, and it runs no queries on your compute, so it adds nothing to your BigQuery bill. Custom data scans are the exception: they read the tables you choose, and their metric queries run in your own Google Cloud project with your credentials.
Add the masthead-agent-tools plugin to your agent, restart it and sign in to the masthead MCP server once. In Claude Code that takes two commands. The Codex and Antigravity steps are in the install section above and in the Masthead docs.
Yes. Interactive clients sign in through Google SSO, while CI/CD and server agents use a service account token. See the authentication guide in the Masthead docs.
Many investigations can be done from chat, but Masthead is still where your data is monitored and configured.
Install the plugin in minutes. Your agent investigates, traces lineage and prepares the fix, and you approve every change.
