Verificate MCP

What Verificate MCP does

The Verificate MCP MCP server reviews AI-generated software work before it enters a codebase. It is intended for coding assistants that need more than syntax or style checks: the service looks for placeholders and mocks in the execution path, invented APIs, unsupported claims of completion, gaming, bypasses, and other implementation risks.

The service exposes four MCP tools. Two return binary decisions: validate_ai_output reviews code or other AI output, while validate_plan checks a design before implementation. analyze_code provides an advisory review without approving or rejecting the input. generate_code combines generation with validation and returns code that has passed the available gates.

How it works

Validation starts with deterministic reality gates. Any gate can veto the result, including checks for mock or placeholder behavior, hallucinated SDK calls, unsubstantiated completion claims, and attempts to game or bypass validation. Content that survives these checks receives a broader review based on ISO/IEC 25010 and MLOps concerns, including performance, scalability, reliability, failure modes, rate-limit implications, and technical debt.

Verdict-producing tools return an approval or rejection together with severity-ranked findings. The analysis tool returns scores and findings but intentionally does not issue a verdict. The generation tool applies the same protection process to newly written code before returning it.

The server also provides gate_my_changes and review_my_plan prompts. These support a validate, fix, and re-validate workflow. The verificate://gates resource describes the deterministic checks, and verificate://example-verdict contains a production rejection example.

Setup and configuration

Verificate MCP MCP server runs as a hosted HTTP service at https://mcp.verificate.ai/mcp, so a local package or process is not required. Add that URL to an MCP client using HTTP transport. Claude Code can be configured with its MCP command, while Cursor, VS Code, and Windsurf can use their client configuration or installation links. A generic configuration uses the server URL and http transport.

The service offers 25 validations per machine without signup, a payment card, or an API key. Continued usage requires an account and token after the free allowance and trial terms. Pass the token as an Authorization: Bearer header in the MCP configuration. The README also documents a 30-day trial and a subsequent $30 monthly price.

Tools and capabilities

• validate_ai_output: returns an approve or reject decision and ranked findings for AI-produced changes.
• validate_plan: evaluates completeness, feasibility, scalability implications, and risk before coding begins.
• analyze_code: reviews existing code with scores and findings, without a binary verdict.
• generate_code: produces code and validates it in the same operation, excluding placeholders and invented APIs.
• Prompts support repeated validation workflows for changes and plans.
• Resources expose gate descriptions and a sample rejection.

The Verificate MCP MCP server can also be used as a standing review rule in assistant instructions or wired into a CI merge gate. The provided material describes the service as hosted; it does not document local execution, language-specific SDKs, or configurable individual gates.