AI Agent Access (Team / Enterprise)¶

gitultra-mcp is an MCP (Model Context Protocol) server so an AI agent -- Claude Desktop, Claude Code, or anything else that speaks MCP -- can query your engineering metrics conversationally, instead of you (or your dashboard) calling the API by hand.

It's a thin layer on top of that same API: every tool call is one HTTP request to your own already-running eng-api instance, using the same credentials you already set up for it. No new database connection, no new data collected -- if you already run eng-api, turning this on is just adding one more container.

Requirements¶

A running eng-api instance (see API Access) -- this has no other dependency
Same Team/Enterprise plan eng-api itself needs

Setup¶

Using the eng-metrics-suite-pro compose bundle, gitultra-mcp is already defined as a service -- set these in .env (mostly reusing what eng-api already needs):

ENG_API_KEY=<same value you already set for eng-api>
GITULTRA_LICENSE_KEY=<your license key, covering "gitultra-mcp">
GITULTRA_MCP_PORT=8100

If your existing license only covers eng-api, ask for a reissue that also includes gitultra-mcp in its product list -- one key can cover both.

Standalone (outside the compose bundle):

docker run --rm -p 8100:8000 \
    -e ENG_API_BASE_URL=http://eng-api:8000 \
    -e ENG_API_KEY=<same value as eng-api's API_KEY> \
    -e GITULTRA_LICENSE_KEY=<your license key> \
    ghcr.io/gitultrahq/gitultra-mcp:latest

Tools¶

One tool per API endpoint, same data, same scoping rules as API Access describes -- gitultra-mcp just forwards whatever you (or the agent) pass through and relays the API's response or error. Every tool accepts either period ("last-week"/"last-month"/"last-quarter") or explicit start/ end, so an agent can say "last week" instead of computing an exact date range.

Tool	Same as
`author_activity`	Author activity
`review_health`	Review health
`repo_trends`	Repo trends
`contributor_concentration`	Contributor concentration
`recency_skew`	Recency skew
`team_repo_activity_mismatch`	Team-to-repo activity mismatch
`contributor_transition_impact`	Contributor departure/reassignment impact
`ticket_lead_time`	Ticket-to-first-commit lead time
`ticket_scope_mismatch`	Ticket scope mismatch
`planning_signature`	Planning signature
`cycle_time`	Cycle time
`deployment_frequency`	Deployment frequency
`lead_time`	Lead time for changes
`change_failure_rate`	Change failure rate
`mttr`	Mean time to restore
`ai_usage`	AI usage
`investment_allocation`	Investment allocation
`author_distribution_trend`	Author distribution trend
`reviewer_distribution_trend`	Reviewer distribution trend
`commits_after_open_distribution_trend`	Commits-after-open distribution trend

investment_allocation is the one tool with no repo/org/team scoping at all, same as its API endpoint -- Jira work items have no repo relationship.

Connecting a client¶

Claude Code¶

claude mcp add --transport http gitultra http://localhost:8100/mcp \
    -H "Authorization: Bearer <your ENG_API_KEY value>"

Claude Desktop¶

Add to your MCP server config:

{
  "mcpServers": {
    "gitultra": {
      "type": "http",
      "url": "http://localhost:8100/mcp",
      "headers": {
        "Authorization": "Bearer <your ENG_API_KEY value>"
      }
    }
  }
}

Getting access¶

Same as API Access -- available on Team and Enterprise plans, see gitultra.com for plan details or to apply for beta access. Full setup details, auth model, and troubleshooting live in gitultra-mcp's own README.