> ## Documentation Index
> Fetch the complete documentation index at: https://docs.gym.scale.com/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP Verifiers

> LLM Judge and rubric claims for tool-use tasks

MCP environments use client-side LLM Judge evaluation to verify agent responses against rubric claims.

## Access

<Note>
  MCP verifiers are only accessible through the Scale Gymnasium Web UI at [gym.scale.com](https://gym.scale.com), not via Docker container APIs.
</Note>

***

## Evaluation Approach

| Component         | Purpose                                               |
| ----------------- | ----------------------------------------------------- |
| **Rubric Claims** | Expected actions/information the agent should produce |
| **LLM Judge**     | Your model evaluates the agent's response             |
| **Verdicts**      | Pass/fail determination per claim                     |

***

## GTFA Claims

Ground Truth Factual Assertions (GTFA) define what must be present in the agent's response.

### What to Include

* **Required information**: Facts that must appear in the response
* **Expected actions**: Tool calls the agent should have made
* **Factual accuracy**: Correct values and calculations
* **Format requirements**: Structure of the expected output

### Example Claims

| Task Type             | Example Claim                                        |
| --------------------- | ---------------------------------------------------- |
| Information retrieval | "Response includes the meeting time of 2:00 PM"      |
| Calculation           | "Response shows total of \$1,234.56"                 |
| Tool use              | "Agent called `send_email` with correct recipient"   |
| Multi-step            | "Agent created calendar event AND sent notification" |

***

## LLM Judge

The LLM Judge evaluates the agent's response against each rubric claim.

### How It Works

1. Agent completes task and produces response
2. Response is sent to LLM Judge with rubric claims
3. Judge evaluates each claim independently
4. Verdicts are aggregated into final score

### Judge Prompt Structure

The judge receives:

* The original task prompt
* The agent's response/trajectory
* Each rubric claim to evaluate

### Verdict Types

| Verdict     | Meaning                                         |
| ----------- | ----------------------------------------------- |
| **Pass**    | Claim is satisfied by the response              |
| **Fail**    | Claim is not satisfied                          |
| **Partial** | Claim is partially satisfied (where applicable) |

***

## Scoring

### Calculation

Final score is typically:

* `1.0` if all claims pass
* `0.0` if any claim fails
* Or proportional based on passed/total claims

### Response Structure

```json theme={null}
{
  "score": 1.0,
  "claims": [
    { "claim": "Agent sent email to correct recipient", "verdict": "pass" },
    { "claim": "Email subject matches requirement", "verdict": "pass" }
  ],
  "message": "All claims verified"
}
```

***

## Task Component Structure

MCP tasks include these verification components:

| Component       | Purpose                                        |
| --------------- | ---------------------------------------------- |
| `ENABLED_TOOLS` | Tools available to the agent                   |
| `PROMPT`        | Task instructions                              |
| `TRAJECTORY`    | Expected sequence of tool calls (ground truth) |
| `GTFA_CLAIMS`   | Required factual claims for verification       |

***

## Best Practices

<Accordion title="Be specific in claims">
  ❌ "Response is correct"

  ✅ "Response includes the customer's order total of \$156.78"
</Accordion>

<Accordion title="Include both positive and negative criteria">
  * ✅ "Response mentions the 3:00 PM meeting"
  * ✅ "Response does NOT include cancelled events"
</Accordion>

<Accordion title="Consider partial completion">
  For complex tasks, define claims at different granularity levels to capture partial success.
</Accordion>

<Accordion title="Test claims with edge cases">
  Verify your claims work correctly with:

  * Correct responses (should pass)
  * Incorrect responses (should fail)
  * Partially correct responses (should behave as expected)
</Accordion>

***

## Example Task Structure

```json theme={null}
{
  "ENABLED_TOOLS": ["calendar_add_event", "email_send", "reminder_create"],
  "PROMPT": "Schedule a meeting with John for Friday at 2pm and send him an email confirmation",
  "TRAJECTORY": [
    { "tool": "calendar_add_event", "args": { "title": "Meeting with John", "time": "Friday 2pm" } },
    { "tool": "email_send", "args": { "to": "john@example.com", "subject": "Meeting Confirmation" } }
  ],
  "GTFA_CLAIMS": [
    "Agent created a calendar event titled 'Meeting with John'",
    "Event is scheduled for Friday at 2:00 PM",
    "Agent sent email to john@example.com",
    "Email subject contains 'Meeting' or 'Confirmation'"
  ]
}
```

***

## Next Steps

<CardGroup cols={2}>
  <Card title="Website Verifiers" icon="globe" href="/deep-dives/verifiers/website">
    State and log checks for web apps
  </Card>

  <Card title="Task Design" icon="clipboard" href="/deep-dives/task-design/overview">
    Create effective MCP tasks
  </Card>
</CardGroup>