widescreen-research: Command for Claude Code

/mcp-dogfood

Enumerate, plan, and execute a full dogfooding pass against an MCP server from within the agent. Produces a machine-updated game state JSON and a RESULTS.md with observations and next steps.

Usage

/mcp-dogfood "$ARGUMENTS" [--run_scenarios=true] [--reasoner=sequentialthinking] [--time_window=15m]

Parameters

• $ARGUMENTS: Target server id or connection name
• --run_scenarios: If true, the agent will execute the planned scenarios (default: true)
• --reasoner: Optional reasoning server to structure scenario planning (e.g., sequentialthinking, clear-thought). Default: sequentialthinking
• --time_window: Optional time window for logs/metrics correlation

Reasoning workflow (sequentialthinking example)

When --reasoner=sequentialthinking is used, the agent plans dogfood scenarios through an explicit chain-of-thought loop using the server's sequentialthinking tool. The workflow looks like this:

1. Initialize thinking context

• Start with thoughtNumber = 1, an initial totalThoughts estimate (e.g., 3), and nextThoughtNeeded = true.

1. Iterative reasoning calls

• For each step, call the tool with:
  • thought (free-form plan/reflection content)
  • thoughtNumber, totalThoughts, nextThoughtNeeded
  • Optional flags to control structure:
    • isRevision, revisesThought to correct prior steps
    • branchFromThought, branchId to explore alternatives in parallel
    • needsMoreThoughts to signal increasing scope mid-run

1. Read back tool output (returned as text JSON)

• Fields include: thoughtNumber, totalThoughts, nextThoughtNeeded, branches, thoughtHistoryLength.
• Adjust counters and decide whether to continue, revise, or branch.

1. Convergence

• Set nextThoughtNeeded = false when the scenario ordering and dependency plan are ready (e.g., entity-creation before dependent operations).

Example call payload (conceptual):

{
  "thought": "Enumerate capabilities; identify create→list→delete dependencies; schedule entity creation first.",
  "thoughtNumber": 1,
  "totalThoughts": 3,
  "nextThoughtNeeded": true
}