ralph-loop-mcp
This server runs the Ralph Loop, an iterative multi-model development workflow where a worker model produces work and a reviewer model reviews it until the reviewer decides to SHIP.
Initialize Ralph Loop sessions with a task and configure worker/reviewer models, providers, max iterations, and cross-model review enforcement.
Manage the iterative workflow: get the current task, submit work with a summary, retrieve submitted work for review, submit SHIP/REVISE decisions with feedback, and get feedback for the next iteration.
Run the complete automated loop in one call using
ralph_loop_run, which handles initialization, orchestration, execution, and state management.Inspect session state: get current status (iteration, phase), view worker/reviewer configuration, and retrieve feedback or completion status.
Reset or clear a session, and block an iteration with a reason when the worker gets stuck.
Persist session state and iteration history under
<workspace>/.ralph/{sessionId}/, supporting multiple concurrent sessions.Support multiple LLM providers (Anthropic, OpenAI, Google, GitHub Copilot, Goose) with optional monitoring-agent fallback parsing of raw outputs.
Allows using GitHub Copilot as worker or reviewer in the Ralph Loop.
Allows using Google's Gemini models as worker or reviewer in the Ralph Loop.
Allows using OpenAI's models (e.g., GPT-4) as worker or reviewer in the Ralph Loop.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@ralph-loop-mcp start a Ralph Loop session to implement JWT auth"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
ralph-wiggum-loop
Multi-model iterative development agent plugin — Ralph Loop with cross-model worker/reviewer architecture and SHIP/REVISE decision engine.
Overview
The Ralph Loop implements a two-phase iterative workflow where a worker model does the work and a reviewer model provides cross-model review, iterating until the reviewer says "SHIP".
This project is a cross-platform Agent Plugin that provides both:
MCP Server Mode — 11 tools over stdio for integration with AI agents
CLI Mode — Run the full loop directly from the command line
Related MCP server: AMS Development MCP Servers
Quick Start
CLI Mode
./bin/ralph-loop-runner.sh "Implement user authentication with JWT tokens"MCP Server Mode
# Run without arguments to start the MCP server
./bin/ralph-loop-runner.shWith a Quality Profile
./bin/ralph-loop-runner.sh "Write a REST API" --profile proDocumentation
Full Documentation — API reference, tools, state management
Deployment Guide — Install and configure
Repository
Source: https://github.com/sheldonrobinson/ralph-wiggum-loop
Agent Plugin Spec: https://agent-plugins.org
MCP Spec: https://modelcontextprotocol.io
License
MIT
Available Tools
10 toolsralph_loop_blockA
Block the current iteration with a reason (creates RALPH-BLOCKED.md). Worker uses this when stuck.
Inputs: reason (required), sessionId.
ralph_loop_get_configB
Get the worker/reviewer model configuration for the session.
Inputs: sessionId.
ralph_loop_get_feedbackB
Get reviewer feedback for the next worker iteration. Returns feedback if REVISE, or completion status if SHIP.
Inputs: sessionId.
ralph_loop_get_statusA
Get current status of the Ralph Loop session including iteration, phase, and state.
Inputs: sessionId.
ralph_loop_get_taskC
Get the current task for the worker phase. Returns the task description.
Inputs: sessionId.
ralph_loop_get_workB
Get the worker's submitted work for the reviewer phase. Returns work output and summary.
Inputs: sessionId.
ralph_loop_initializeB
Initialize a new Ralph Loop session with a task. Starts the iterative worker/reviewer cycle.
Inputs: task (required), sessionId, workerModel, maxIterations, +4 more.
ralph_loop_resetA
Reset/clear a Ralph Loop session, removing all state files.
Inputs: sessionId.
ralph_loop_submit_reviewB
Submit review decision (SHIP or REVISE) with feedback from the reviewer phase.
Inputs: decision (required), iteration (required), feedback, sessionId.
ralph_loop_submit_workA
Submit work results and summary from the worker phase. Transitions to review phase.
Inputs: work (required), summary (required), iteration (required), sessionId.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 10 tools
Each tool has a distinct purpose: initialization, work submission, review feedback, status, etc. No two tools have overlapping functionality.
All tools follow a consistent verb_noun pattern with the ralph_loop_ prefix in snake_case, making them predictable and logical.
With 10 tools, the set covers the full worker/reviewer cycle without being bloated or insufficient.
Covers all major phases: initialize, work, review, block, reset, and status. Minor gap: no explicit tool to list all feedback history, but core workflow is complete.
Maintenance
Related MCP Connectors
One MCP server for all your MCP servers: code mode, per-tool permissions, approvals and audit.
MCP Server for an Agent Task Marketplace
MCP server for generating rough-draft project plans from natural-language prompts.
MCP server for Product Management
Related MCP Servers
- AlicenseAqualityCmaintenanceAn MCP server that turns independent AI agents into a coordinated engineering team with shared task board, context, review loop, and enforced plan-implement-review-iterate workflow.24MIT
- AlicenseNot gradedqualityDmaintenanceMCP servers for real-time development monitoring and interactive pair programming.1MIT

convergeqa-mcpofficial
AlicenseBqualityDmaintenanceMCP servers for multi-model document review with critique/iterate and compare/due-diligence tools, using public verification receipts.16MIT- AlicenseNot gradedqualityDmaintenanceMCP server for creating, validating, and reviewing implementation plans in GitHub Copilot Chat, with visual review interface and MongoDB persistence.MIT