yamlr mcp

Start the Yamlr Model Context Protocol (MCP) server.

Why

MCP is the standard protocol for connecting AI agents to precision tools. By running Yamlr as an MCP server, you enable AI assistants (like Cursor, Claude, or custom DevOps agents) to perform surgical Kubernetes manifest analysis and healing with 100% deterministic safety.

Usage

bash
yamlr mcp [OPTIONS]

Options

--transport (default: stdio)

The transport protocol to use for communication.

• stdio: standard input/output (recommended for IDE integrations).
• http: (Not yet implemented in v1.1).

--port (default: 8000)

The port to listen on when using HTTP transport.

MCP Tools Exposed

🔍 scan_manifest

Performs a deep safety audit of a directory or single manifest.

• Input: path (string)
• Output: A safety report containing:
  • Rule ID: Unique identifier for the violation.
  • Rationale: An architectural explanation of why the fix is necessary.
  • Impact Description: The real-world consequence of ignoring the issue.
  • Risk Level: High/Medium/Low assessment.

🔧 heal_manifest

Applies surgical repairs to Kubernetes manifests.

• Input: path (string), rules (optional list of Rule IDs to filter).
• Output: A transaction receipt showing exactly what was changed and a SHA-256 DNA proof of intent preservation.

❓ why

Provides deep architectural context for any Yamlr Rule ID.

• Input: rule_id (string)
• Output: A comprehensive explanation of the security or operational logic behind the rule.

IDE Integration Example (Cursor/Claude Desktop)

To connect Yamlr to your AI agent, add the following to your MCP configuration:

json
{
  "mcpServers": {
    "yamlr": {
      "command": "yamlr",
      "args": ["mcp"]
    }
  }
}

Safety Guardrails

• Read-Only by Default: The scan_manifest tool never modifies files.
• DNA Receipts: Every heal_manifest call generates a cryptographic proof that nothing outside the intended fix was altered.
• Deterministic: Unlike raw LLM suggestions, Yamlr's MCP tools are strictly deterministic and schema-validated.

🛡 Universal Expert Flags

Available across primary commands:

• --timing: [PERFORMANCE] Output surgical execution timing breakdown.
• --timing-file <FILE>: [PERFORMANCE] Save performance profiling data to a JSON file.
• --verbose: Full audit logs and engine stages.
• -q, --quiet: Minimal output mode (exit codes only).