PV-Bhat/vibe-check-mcp-server

Vibe Check is a tool that provides mentor-like feedback to AI Agents, preventing tunnel-vision, over-engineering and reasoning lock-in for complex and long-horizon agent workflows. KISS your over-eager AI Agents goodbye! Effective for: Coding, Ambiguous Tasks, High-Risk tasks

503 ★66 forksTypeScriptUpdated 1d ago

What you need to know

A plug-and-play agent oversight server that acts as an AI meta-mentor, interrupting pattern inertia with research-backed Chain-Pattern Interrupts (CPI) to keep agents on the minimal viable path and prevent Reasoning Lock-In.

Install

npx -y @pv-bhat/vibe-check-mcp start --stdio
npx -y @pv-bhat/vibe-check-mcp start --http --port 2091
bash scripts/docker-setup.sh

Usage

  • Add the vibe-check-mcp entry to your MCP client config (Claude Desktop, Cursor, Windsurf, VS Code) with "command": "npx", "args": ["-y", "@pv-bhat/vibe-check-mcp", "start", "--stdio"].
  • Set a provider API key in .env (GEMINI_API_KEY default, or OPENAI_API_KEY / OPENROUTER_API_KEY / ANTHROPIC_API_KEY) and optionally DEFAULT_LLM_PROVIDER.
  • Instruct your agent in its system prompt to call vibe_check after planning and before major actions, and optionally log mistakes with vibe_learn.
  • Use the session-constitution tools (update_constitution / reset_constitution / check_constitution) to enforce per-session rules such as "no external network calls".

Key features

  • CPI adaptive interrupts: phase-aware prompts that challenge assumptions at risk inflection moments (research: nearly doubles success ~27%->54%, halves harmful actions ~83%->42%).
  • Multi-provider LLM support: Gemini 3.6, Claude 5, GPT-5.6, and OpenRouter, configurable per call with modelOverride.
  • History continuity: summarizes prior advice when sessionId is supplied; optional vibe_learn logs mistakes for future reflection.
  • Client installers (idempotent, tagged "managedBy") for Claude Desktop, Cursor, Windsurf, and VS Code, plus HTTP hardening (CORS loopback, Host validation, body cap).

Best for

Adding a research-backed reflection/safety checkpoint layer in front of autonomous coding agents that tend to over-engineer or go off-track.

Caveats

  • Project is in maintenance mode: active feature development has ended; only maintenance patches (security and bug fixes) are published as of v2.9.0.
  • Requires Node >=20.
  • HTTP mode on non-loopback hosts needs MCP_ALLOWED_HOSTS set or requests get HTTP 403.
Platforms: macOS · Linux · WindowsClients: Claude Desktop · Cursor · Windsurf · Visual Studio Code
Documentation ↗

Reviewed 2026-08-07

Topics

agentic-aiagentic-workflowai-agentschain-of-thoughtcpierror-handlingmcpmcp-servermodel-context-protocolrlivibe-codingworkflow-automation
View on GitHub ↗
Stars
503★
Forks
66
Language
TypeScript
License
MIT
Created
2025-03-22
Last push
2026-08-05