M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
V

Vibe Check MCP

Community
KISS overzealous agents goodbye: plug & play agent oversight tool.
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #74 of 438
Stars
★ 502 Very popular
Transport
stdio (local process) · Streamable HTTP
Runtime
Node.js >=20
Credentials
Optional API key
License
MIT
Last commit
Tools
5
63FMRS · C

Vibe Check MCP offers a unique metacognitive oversight layer with research-backed CPI methodology. It is easy to integrate (npx + MCP clients) and supports multiple LLM providers. However, the server is in maintenance mode (only security/bug fixes), and reliance on external LLMs introduces cost and privacy considerations.

Strongest · Setup experience 14/20 Weakest · Reliability 10/20

Reliability
10/20
Security and permissions
12/20
Maintenance
13/20
Documentation
14/20
Setup experience
14/20
Why each score
Reliability 10/20
Evidence: server.json and README consistently declare stdio and HTTP transports, and the tool list includes vibe_check, vibe_learn, and constitution-related tools. However, advanced features like vibe_learn and constitution are mentioned in README, but their actual tool list implementation is not verified in the provided source. Maintenance mode indicates no active bug fixes. No actual execution evidence (CI logs, test results) is provided to confirm the server starts cleanly and completes MCP handshake. Therefore, reliability score is capped at 12 due to lack of direct execution verification. Deductions: cannot verify tool declarations match real behavior; no CI logs or test run evidence.
Security and permissions 12/20
Evidence: README details HTTP transport hardening (CORS defaults to loopback, Host header validation, request body size limits). API keys are managed via environment variables and marked as secrets. No malicious behavior, credential theft, or irreversible operations by default. However, this static review does not execute code, so these security controls are unverified. Also, prompt injection risk exists (input sanitization not implemented, as per README roadmap). Score 12 because security measures are documented but unverified, and some risks (e.g., prompt injection) remain unmitigated. Deductions: no actual code verification; security controls are only documented; input sanitization not implemented.
Maintenance 13/20
Evidence: Repository is active with 503 stars and 3 open issues, and a maintenance mode declaration (only security/bug fixes after v2.9.0). MIT license is present, and SECURITY.md provides a clear reporting channel. Dependency updates are noted in v2.9.0 changelog. However, maintenance mode indicates feature development has stopped, which may affect long-term maintenance. Also, no detailed commit frequency or release dates are provided. Score 13 because maintenance is active with license and security policy, but maintenance mode limits future development. Deductions: no specific evidence of commit/release frequency.
Documentation 14/20
Evidence: README is extremely comprehensive, covering installation, providers, models, transports, client configurations, Docker, security, roadmap, FAQ, etc. There are links to many docs like docs/architecture.md, docs/integrations/cpi.md, docs/changelog.md. It includes environment variable tables, configuration options, and troubleshooting tips. However, some doc links may be empty (e.g., docs/TESTING.md link appears non-functional), and limitations/costs are not clearly disclosed. Score 14 because documentation is high quality but has some dead links and missing cost info. Deductions: some doc links may be invalid; missing explicit cost discussion.
Setup experience 14/20
Evidence: Provides npx quickstart commands, supports stdio and HTTP transports, and includes client config examples (Claude Desktop, etc.). Requires Node >=20 and has an automatic Docker setup script. However, no actual verified CI tests or installation test results are provided, so score capped at 15. Score 14 because installation steps are clear, but requiring Node version and external API keys may add complexity. Deductions: no verification of installation process; environment variables may be many.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessUses the network

Best for

  • Developers and teams building AI agents that need extra oversight to prevent reasoning lock-in and over-engineering.
  • Users working with MCP clients like Claude Desktop, Cursor, Windsurf, or VS Code who want long-horizon workflows to stay on track.
  • Developers seeking research-backed reliability improvements for agents.

Not for

  • Simple or fast tasks where an external oversight call adds unwanted latency and cost.
  • Scenarios where you cannot or do not want to rely on external LLM API calls (requires at least one provider key).
  • Projects requiring active feature development or official support, as the server is in maintenance mode.

Required permissions

  • Requires at least one LLM provider API key (Gemini, OpenAI, OpenRouter, or Anthropic) to function.
  • HTTP mode opens a network port; recommend restricting to loopback for security.
  • May modify client configuration files (e.g., Claude Desktop, Cursor, Windsurf, VS Code) during installation to add the server entry.

Risks and side effects

  • Sending prompts and context to external LLM APIs; ensure sensitive data is not exposed.
  • HTTP mode with non-loopback access increases attack surface; v2.9.0 adds host validation and CORS restrictions but misconfiguration could cause outages.
  • Over-frequent `vibe_check` calls can slow down agent execution and increase costs; optimal dose is ~10-20% of steps.

Setup

Before you start

Runtime：Node.js >=20

GEMINI_API_KEY optionalsecret Google AI Studio (Gemini) API key; default provider, at least one provider key is required.
OPENAI_API_KEY optionalsecret OpenAI API key; optional provider.
OPENROUTER_API_KEY optionalsecret OpenRouter API key; optional provider, and DEFAULT_MODEL is required when using openrouter.
ANTHROPIC_API_KEY optionalsecret Anthropic API key; optional provider.
Other optional settings (7)
DEFAULT_LLM_PROVIDER optional Provider used when no modelOverride is supplied: gemini | openai | openrouter | anthropic (default: gemini).
DEFAULT_MODEL optional Model of DEFAULT_LLM_PROVIDER to use when no modelOverride is supplied; leave unset for the per-provider default.
CORS_ORIGIN optional HTTP mode only: comma-separated browser origin allowlist, defaults to loopback origins only.
MCP_ALLOWED_HOSTS optional HTTP mode only: Host header allowlist for DNS-rebinding protection, defaults to localhost/127.0.0.1/::1.
MCP_MAX_BODY_SIZE optional HTTP mode only: JSON body size cap, default 100kb.
MCP_TRANSPORT optional Transport selection used when no explicit flag is given, default stdio.
MCP_HTTP_PORT optional HTTP mode port, default 2091.
  1. Ensure Node.js >= 20.
  2. Run npx -y @pv-bhat/vibe-check-mcp start --stdio for stdio or --http --port 2091 for HTTP.
  3. Obtain a Gemini API key (default provider) or optionally keys for OpenAI, OpenRouter, or Anthropic.
  4. Add the server entry to your MCP client config (e.g., Claude Desktop, Cursor, Windsurf, VS Code).
  5. Optionally set environment variables DEFAULT_LLM_PROVIDER and DEFAULT_MODEL to choose the provider and model.
claude_desktop_config.json 
{
  "mcpServers": {
    "vibe-check-mcp": {
      "command": "npx",
      "args": ["-y", "@pv-bhat/vibe-check-mcp", "start", "--stdio"]
    }
  }
}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

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

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

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

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

Start the server via `npx -y @pv-bhat/vibe-check-mcp start --stdio` from an MCP client (Claude Desktop, Cursor, etc.) and confirm vibe_check appears in the client's tool list; calling vibe_check with a goal and plan and receiving mentor-style feedback confirms the connection.

Troubleshooting

  1. Check Node.js version (>=20) if startup fails.
  2. Ensure at least one API key is set; if keeping Gemini as default, set `GEMINI_API_KEY`.
  3. If HTTP requests are rejected, verify `MCP_ALLOWED_HOSTS` includes your hostname or set to `*` to disable the check.
  4. If encountering rate limits or errors, review model compatibility and consider using `modelOverride`.
  5. Use `npx -y @pv-bhat/vibe-check-mcp doctor` (if available) to diagnose configuration issues.

Things to try

Once connected, you can ask your AI assistant things like:

  • Call vibe_check after planning this feature to challenge my assumptions and check for over-engineering
  • Log this resolved bug and its fix with vibe_learn so we can reflect on it later
  • Set session rules for this sessionId: no external network calls and write unit tests before refactoring
  • Show me the effective constitution rules for the current session

Tools 5

vibe_check read-only
Challenges assumptions and prevents tunnel vision after planning and before major actions.
vibe_learn writes
Logs mistakes, preferences, and successes for future reflection and self-improvement.
update_constitution writes
Sets or merges session rules for a given sessionId that the CPI layer will enforce.
reset_constitution destructive
Clears session rules for a given session.
check_constitution read-only
Inspects the effective rules for a given session.

Use cases

Coding tasks: check plans before writing code to avoid over-engineering.
Ambiguous tasks: help agents navigate unclear requirements and stay aligned.
High-risk tasks: inject reflective pauses before irreversible actions to reduce harmful behavior.

Supported clients

Claude Desktop
Cursor
Windsurf
VS Code

Listed from the project's documentation, not tested by this site.

Overview

Vibe Check MCP is a lightweight server implementing Anthropic's Model Context Protocol (MCP). It acts as an AI meta-mentor for your agents, interrupting pattern inertia with Chain-Pattern Interrupts (CPI) to prevent Reasoning Lock-In (RLI). It keeps agents on the minimal viable path, escalating complexity only when evidence demands it. The server invokes a second LLM to provide metacognitive feedback, helping agents pause and reflect before major actions, improving alignment, common sense, and safety. Research-backed: in evaluations, agents using Vibe Check improved success by ~27% and halved harmful actions (-41%).

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision ed8452f2d626 Data synced 2026-10-19 Read the FMRS scoring method