Vibe Check MCP

Research-backed mentor layer that interrupts agent pattern inertia to prevent over-engineering and reasoning lock-in.

Unverified

stdio (local)

API key

TypeScript

View repo 492 Website npm

Add to your client

Copy the config for your MCP client and paste it into its config file.

Install / run

npx -y @pv-bhat/vibe-check-mcp start --stdio

Paste into ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "vibe-check-mcp": {
      "command": "npx",
      "args": [
        "-y",
        "@pv-bhat/vibe-check-mcp",
        "start",
        "--stdio"
      ]
    }
  }
}

Step-by-step guides: Add to Claude Desktop · Add to Cursor · Add to Windsurf

Before you start

Node.js >= 20
An LLM provider API key (GEMINI_API_KEY by default; OPENAI_API_KEY, ANTHROPIC_API_KEY, or OPENROUTER_API_KEY also supported)
An MCP-aware client (Claude Desktop, Cursor, Windsurf, VS Code, etc.)

About Vibe Check MCP

Vibe Check MCP keeps agents on the minimal viable path and escalates complexity only when evidence demands it. It pairs a metacognitive signal layer (surfacing traits, uncertainty, and risk scores) with Chain-Pattern Interrupts (CPI) so agents pause when risk spikes, before resuming under an intervention policy. Integrating vibe_check calls into an agent's system prompt — especially before irreversible actions — significantly improves alignment and common-sense behavior. The server invokes a second LLM (Gemini, OpenAI, Anthropic, or OpenRouter) to generate the feedback.

Tools & capabilities (5)

vibe_check

Challenge assumptions and prevent tunnel vision; invokes a second LLM to give metacognitive feedback on the agent's plan. Pass the full user request and current plan; supports sessionId for history continuity.

vibe_learn

Optionally capture mistakes, preferences, and successes to build a history for future reflection and self-improvement.

update_constitution

Set or merge a per-session rule set (constitution) that the CPI layer will honor, keyed by sessionId.

reset_constitution

Clear the constitution rules for a given session.

check_constitution

Return the effective constitution rules currently in force for a given session.

When to use it

Preventing over-engineering by interrupting agents on the minimal viable path
Adding research-backed oversight to long-horizon or high-risk agent workflows
Sanity-checking an agent's plan before irreversible actions
Enforcing per-session rules (e.g. no external network calls, tests before refactors) via a session constitution
Logging and reflecting on past mistakes to improve future agent runs

Security notes

Requires an LLM provider API key (Gemini by default; OpenAI, Anthropic, or OpenRouter also supported) supplied via environment variables (e.g. GEMINI_API_KEY) or a .env file. The server invokes a second LLM, so prompts/context are sent to the configured provider. A CI-based security scan runs npm audit and source pattern checks; see SECURITY.md. v2.8.0+ updated the MCP SDK to 1.26 to address a cross-client data-leakage fix and resolved 14 npm audit vulnerabilities.

Vibe Check MCP FAQ

What problem does Vibe Check solve?

LLMs can confidently follow flawed plans and spiral into over-engineering or misalignment (pattern inertia / reasoning lock-in). Vibe Check provides an external nudge via short reflective pauses, improving reliability and safety.

Which LLM providers are supported?

Gemini (default), OpenAI, Anthropic (including Anthropic-compatible endpoints), and OpenRouter. Set DEFAULT_LLM_PROVIDER and the matching API key.

Which transports are supported?

Both STDIO and streamable HTTP. Transport resolution follows explicit flags (--stdio/--http) → MCP_TRANSPORT → default stdio. For HTTP, the default port is 2091.

Is the project actively developed?

It is in maintenance mode (latest release v2.8.1) with only security and bug-fix patches; it remains fully functional under the MIT license and welcomes community forks.

#agent-oversight #metacognition #cpi #reflection #ai-agents #alignment #vibe-coding #llm

Alternatives to Vibe Check MCP

Compare all alternatives →

Playwright MCP Server

Developer Tools

24k

Microsoft's official browser-automation MCP using Playwright's accessibility tree (no vision model).

Featured

Verified

stdio (local)

No auth

TypeScript

12 tools

Updated 21 days agoRepo

Context7 MCP Server

Developer Tools

28k

Up-to-date, version-specific library documentation injected into your coding agent.

Verified

stdio (local)

API key

TypeScript

2 tools

Updated 25 days agoRepo

codebase-memory-mcp

Developer Tools

16k

Code intelligence engine that indexes repos into a persistent knowledge graph for AI coding agents.

Unverified

stdio (local)

No auth

14 tools

Updated 11 hours agoRepo

Compare Vibe Check MCP with:

vs Playwright MCP Server vs Context7 MCP Server vs codebase-memory-mcp vs Serena