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

# Agents and MCP

> Connect compatible AI clients to Cloudeval tools, select review profiles, and automate safely with evidence boundaries.

Cloudeval exposes project context and review tools to compatible clients. The client supplies the interaction surface; Cloudeval supplies scoped tools, project evidence, and evaluation contracts.

<Columns cols={2}>
  <Card title="Connect an MCP client" icon="plug" href="/agents/mcp-client-setup">
    Configure Codex, Cursor, Claude Code, VS Code, or a generic MCP host.
  </Card>

  <Card title="Choose an Agent Profile" icon="bot" href="/agents/profiles">
    Select architecture, cost, triage, remediation, visual, scripting, change-review, evidence, or security behavior.
  </Card>

  <Card title="Review automation rules" icon="shield-check" href="/agents/automation-rules">
    Understand context, credentials, evidence boundaries, failure behavior, and safe automation.
  </Card>

  <Card title="Use agent context files" icon="files" href="/agents/context-files">
    Use the public machine-readable product context and search surfaces.
  </Card>
</Columns>

<Warning>Agent output is not a substitute for missing evidence. When source, report, or cloud evidence is stale or unavailable, the agent should report the gap instead of inventing a conclusion.</Warning>

## Recommended path

<Steps>
  <Step title="Connect one client">
    Start with the MCP setup for the client your team already uses.
  </Step>

  <Step title="Choose the narrowest profile">
    Match the profile to the decision you need instead of exposing unnecessary tools.
  </Step>

  <Step title="Verify source and freshness">
    Confirm the selected project, source revision, report state, and evidence timestamp before acting on a recommendation.
  </Step>
</Steps>

## Related pages

* [CLI workflows](/cli/workflows)
* [Evaluation methodology and sources](/trust/evaluation-methodology)
* [Data handling and AI boundaries](/trust/data-and-ai)