For AI agents

Let your agent raise the score

Agnes tells you what is wrong with your frontend code. Connect your AI agent to Agnes and it can read the report, work out which fixes gain the most points, and write a plan before it touches your code.

How it works

  1. 1. Create a token

    In the Agnes app, go to Settings → API tokens and create a token. It can only read, and only your team's reports.

    Open the app's settings →

  2. 2. Connect your agent

    Add the Agnes MCP server to your agent with the token, using one of the setups below.

  3. 3. Ask for a plan

    Ask your agent to plan how to raise the score. It reads the latest report, ranks the fixes by the points they gain, and shows you the plan first.

    “Make a plan to raise the Agnes score of this repository.”

Connect your agent

Agnes runs an MCP server at the address below. Replace <your token> with the token from the app.

Claude Code

claude mcp add --transport http agnes https://app.agnes.dev/api/mcp --header "Authorization: Bearer <your token>"

Cursor

In ~/.cursor/mcp.json

{
  "mcpServers": {
    "agnes": {
      "url": "https://app.agnes.dev/api/mcp",
      "headers": {
        "Authorization": "Bearer <your token>"
      }
    }
  }
}

VS Code

In .vscode/mcp.json

{
  "servers": {
    "agnes": {
      "type": "http",
      "url": "https://app.agnes.dev/api/mcp",
      "headers": {
        "Authorization": "Bearer ${input:agnes-token}"
      }
    }
  },
  "inputs": [
    {
      "type": "promptString",
      "id": "agnes-token",
      "description": "Agnes API token",
      "password": true
    }
  ]
}

Other clients

URL:    https://app.agnes.dev/api/mcp
Header: Authorization: Bearer <your token>
Transport: Streamable HTTP

Add the skill (optional)

The skill gives your agent the full workflow: find the project, read the report, rank the fixes, check each finding in your code, and write a plan to approve before any change. It works in Claude Code, Claude and other agents that support skills.

Download the skill (zip) SKILL.md

Or install it for Claude Code with one command:

mkdir -p ~/.claude/skills/agnes-improve-score && curl -fsSL https://agnes.dev/skills/agnes-improve-score/SKILL.md -o ~/.claude/skills/agnes-improve-score/SKILL.md

What your agent can do

list_projects
The projects of your team, with their latest score.
get_report
The overall score, the score per category, the findings per category and which checks did not run.
rank_fixes
The rules whose findings cost the most points, with how many points fixing each would gain.
get_findings
The findings themselves, 50 per page, by rule, category, severity or folder, with file, line and a link to the rule.
get_scoring
How findings turn into scores, so the agent can explain its plan.

Safe by design

  • Read-only

    The tools only read reports. Your agent cannot change projects, settings or anything else in Agnes.

  • Your team only

    A token works for the team it was created in, as long as you are a member of it.

  • Revoke at any time

    Revoke a token in Settings and it stops working straight away. The app shows when each token was last used.

  • A plan before changes

    The workflow asks your agent to write a plan and wait for your approval before it changes any code.