crash
MCP server for structured and efficient reasoning with step validation, branching, and revisions.
claude mcp add --transport stdio nikkoxgonzales-crash-mcp npx -y crash-mcp
How to use
CRASH is an MCP server that enables structured, adaptive reasoning with trackable steps, confidence tracking, and branching for exploring alternative solutions. It outputs steps in a repeatable format and supports multiple output formats (console, JSON, Markdown). To use it, simply run the server via the provided command (npx -y crash-mcp) or adapt to your preferred client (e.g., Windows wrapper, Docker, Bun, Deno) as described in the README. Once running, you can interact with the server by issuing step-by-step reasoning prompts, retrieving structured plan steps, and leveraging the revision and branching features to refine results. The system supports strict and flexible validation modes and can be configured with environment variables to control behavior like history size, session timeouts, and output format. Tools available include structured thinking with dependencies, session management, multiple output formats, and a configurable validation strategy. CRASH is especially useful when complex problems require methodical, verifiable reasoning across multiple possible paths.
How to install
Prerequisites:
- Node.js (recommended latest LTS)
- Basic familiarity with npm or npx
-
Install the MCP server package globally or run via npx (no local install required):
npm install crash-mcp
Or directly run:
npx crash-mcp
-
If you want to run via Docker (optional):
Create a Dockerfile or use the provided example:
FROM node:18-alpine WORKDIR /app RUN npm install -g crash-mcp CMD ["crash-mcp"]
Then build/run: docker build -t crash-mcp . docker run -i --rm crash-mcp
-
Quick setup ( JSON config for MCP-compatible clients):
{ "mcpServers": { "crash": { "command": "npx", "args": ["-y", "crash-mcp"] } } }
-
Optional environment variables (examples):
- CRASH_STRICT_MODE, MAX_HISTORY_SIZE, CRASH_OUTPUT_FORMAT, CRASH_SESSION_TIMEOUT, CRASH_MAX_BRANCH_DEPTH
- Optional runtimes (alternative): Bun, Deno, etc. See README for details on other runtimes and client configurations.
Additional notes
Tips and notes:
- You can enable environment-based configuration by including an env block in your MCP config, adjusting strict mode, history limits, and output format as needed.
- For Windows users, a cmd wrapper is provided to execute NPX-based runs.
- If you run CRASH behind a proxy or in restricted environments, ensure network access for fetching the npm package when using npx.
- Use the JSON config examples to integrate with clients like Claude Code, Cursor, VS Code, Windsurf, JetBrains, or other MCP-enabled tools.
- When exploring branches, track dependencies between steps and set appropriate branch_depth limits to prevent unbounded exploration.
Related MCP Servers
penpot
Penpot MCP server
Remote
A type-safe solution to remote MCP communication, enabling effortless integration for centralized management of Model Context.
git
An MCP (Model Context Protocol) server enabling LLMs and AI agents to interact with Git repositories. Provides tools for comprehensive Git operations including clone, commit, branch, diff, log, status, push, pull, merge, rebase, worktree, tag management, and more, via the MCP standard. STDIO & HTTP.
mcp -odoo
A Model Context Protocol (MCP) server that enables AI assistants to securely interact with Odoo ERP systems through standardized resources and tools for data retrieval and manipulation.
boilerplate
TypeScript Model Context Protocol (MCP) server boilerplate providing IP lookup tools/resources. Includes CLI support and extensible structure for connecting AI systems (LLMs) to external data sources like ip-api.com. Ideal template for creating new MCP integrations via Node.js.
notification
A Model Context Protocol server that allows AI agents to play a notification sound via a tool when a task is completed.