Model Context Protocol moderate risk

Vibe Check

Vibe Check MCP is a research-backed AI meta-mentor that uses Chain-Pattern Interrupts (CPI) to keep agents aligned, reflective, and safe.

Connections & Capabilities

Connects To

GitHubDocker

Capabilities

readwriteexec

Quickstart

Install

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

Config

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

Exposed MCP Tools (3)

high

update_constitution

Merges or sets the rule set for a specific session.

Modifies session rules, potentially impacting agent behavior and access.

high

reset_constitution

Clears the session rules for a specific session.

Removes all session rules, potentially removing critical restrictions.

safe

check_constitution

Returns the effective rules for a specific session.

Read-only operation, no side effects.

Safety Assessment

Vibe Check MCP enhances agent safety by providing metacognitive feedback and interrupting flawed reasoning. It is relatively safe when used with strong API key management and well-defined session constitutions. Risks increase if API keys are exposed or constitutions are misconfigured, potentially leading to unintended consequences or security breaches.

Uses a second LLM for metacognitive feedback, adding a layer of oversight.
Implements Chain-Pattern Interrupts (CPI) to prevent reasoning lock-in.
Supports session constitutions to enforce rules per session.
Offers optional vibe_learn for self-improvement through logging mistakes and fixes.
Provides history continuity by summarizing prior advice when a sessionId is supplied.

Relies on external LLM providers (Gemini, OpenAI, Anthropic, OpenRouter) which introduces dependency risks.
Requires API keys for LLM providers, potentially exposing secrets if not managed correctly.
Session constitutions, if not properly configured, could lead to unintended restrictions or bypasses.
HTTP transport exposes an endpoint that needs proper security measures to prevent unauthorized access.
Write operations are possible via constitution updates, posing a risk if exploited.