Vibe Check

Vibe Check is an MCP server that provides an automated oversight and sanity-check layer for AI agents to prevent reasoning lock-in, scope creep, and compounding errors. Acting as a meta-mentor layer, the server connects primary agent workflows to secondary language model evaluators across providers like Gemini, Claude, GPT, and OpenRouter. It is designed for vibe coders, software developers, and agent system designers who need to keep autonomous LLMs aligned with minimal viable implementations. By introducing Chain-Pattern Interrupts, Vibe Check surfaces traits, uncertainty indicators, and risk scores before irreversible actions or heavy refactors take place. The server supports continuity through session-based history tracking, mistake logging with optional reflection, and per-session rule enforcement via custom constitution tools. It runs locally through stdio or over an HTTP endpoint, allowing development environments and clients to request immediate second-opinion assessments during active code generation and tool usage.

Category: AI & LLM Tooling

Tags: agent-framework, alignment, code-quality, error-prevention

Visit Vibe Check

How to install and configure Vibe Check

To set up and connect Vibe Check to an MCP client, follow these steps: 1. Ensure your system has Node.js version 20 or higher installed. 2. Configure your MCP client (such as Claude Desktop or Cursor) to execute the server via standard I/O. 3. Add the following entry to your client configuration file: json { "mcpServers": { "vibe-check-mcp": { "command": "npx", "args": ["-y", "@pv-bhat/vibe-check-mcp", "start", "--stdio"] } } } 4. Alternatively, run the HTTP server directly in your terminal: bash npx -y @pv-bhat/vibe-check-mcp start --http --port 2091 5. Restart your MCP client and verify the server connects.

What you can do with Vibe Check

• Review proposed software refactors before execution to prevent unnecessary over-engineering and retain minimal viable paths. - Intercept automated file modifications and command runs with risk scoring prior to irreversible systemic disk changes. - Enforce project constraints and safety guidelines across multi-turn sessions using session-specific constitutional rule checks. - Capture reasoning errors and failed execution paths with historical logging to guide subsequent agent reflection turns. - Evaluate complex architectural planning steps using multi-provider secondary LLMs to eliminate cognitive lock-in.

Key facts

• https://github.com/PV-Bhat/vibe-check-mcp-server
• AI & LLM Tooling, Developer Tools & Code Intelligence
• agent-framework, alignment, code-quality, error-prevention

Part of MCP Servers

How do I install Vibe Check?

You can run Vibe Check directly without installation by using npx with Node version 20 or higher. Add npx with the arguments -y, @pv-bhat/vibe-check-mcp, start, and --stdio to your MCP client configuration file, or launch an HTTP instance using the --http and --port flags.

What can Vibe Check do?

Vibe Check introduces meta-cognitive sanity checks into agent workflows. It interrupts pattern inertia using Chain-Pattern Interrupts, surfaces uncertainty and risk scores, reviews planned agent actions, maintains advice continuity across sessions, logs past errors for learning, and applies configurable per-session rules via a lightweight session constitution.

Which MCP clients work with Vibe Check?

Vibe Check works with any standard MCP-compliant client that supports standard I/O (STDIO) or HTTP transports, including Claude Desktop, Cursor, and Windsurf.

Is Vibe Check open source?

Yes, Vibe Check is open-source software distributed under the MIT license on GitHub. The project is currently maintained in maintenance mode, accepting community contributions, security fixes, and bug patches.