Verificate MCP Quickstart

What VerificateAI/verificate-mcp-quickstart does

VerificateAI/verificate-mcp-quickstart MCP server is a hosted review service for AI-generated software changes. It is intended for the 💻 Developer Tools category and focuses on finding defects that can look acceptable in a demo but fail in production. Its checks include mock or placeholder behavior in the execution path, nonexistent SDK or API calls, tests designed to pass without proving behavior, bypass patterns, and claims that work is complete without supporting evidence.

The service separates hard gates from advisory analysis. Deterministic checks run first and can reject an output. Code that passes those checks receives a broader review covering areas such as performance, scalability, reliability, and technical debt. Results include a score and severity-ranked findings when applicable.

How it works

The hosted endpoint is https://mcp.verificate.ai/mcp and uses HTTP transport. An MCP-compatible client sends code, a plan, or a related review request to the service. The response depends on the selected tool:

• validate_ai_output returns an approve/reject decision and findings for AI-written code.
• validate_plan evaluates a plan or design before implementation, including completeness, feasibility, scalability implications, and risk.
• analyze_code provides scores and findings for existing code but intentionally does not issue a verdict.
• generate_code produces code and checks it through the same protection engine before returning it.

The server also exposes 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, while verificate://example-verdict contains a production rejection example.

Setup and configuration

VerificateAI/verificate-mcp-quickstart MCP server requires no local installation. Add the hosted URL to an MCP client using HTTP transport. Claude Code can register it with claude mcp add --transport http verificate https://mcp.verificate.ai/mcp. For clients using JSON configuration, define the server URL and set the transport to http.

The README gives ~/.cursor/mcp.json as the Cursor location and ~/.codeium/windsurf/mcp_config.json for Windsurf. It also provides installation links for Cursor and VS Code, and notes that LM Studio and Goose buttons are available from the Verificate website. Each machine receives 25 free validations without an account, card, or key. Continued use requires a signup and token; the documented trial lasts 30 days, after which the listed price is $30 per month.

Tools and capabilities

Use validate_ai_output as a merge gate for substantive AI-generated changes. Use validate_plan before code exists, particularly when checking whether a design is complete and feasible. Choose analyze_code when you need investigation rather than a binary decision, such as examining hot paths, rate-limit calculations, failure modes, or inherited technical debt. generate_code combines generation and validation.

The repository also documents standing instructions for an agent: validate a change before presenting it as complete, fix rejected findings, and avoid claiming tests or deployment succeeded without proof. The service can therefore be used interactively through an assistant or incorporated into a merge-gating workflow.

Limitations and notes

The service is hosted rather than self-installed, so use depends on access to its HTTP endpoint. The free allowance is limited to 25 validations per machine. analyze_code is advisory and does not return approve/reject, unlike the two validation tools. The documented client setup names Claude Code, Cursor, Windsurf, VS Code, LM Studio, Goose, and generic MCP clients; individual client configuration details can differ.

The repository reports measured results for specific adversarial examples, including reward-gaming tests and hallucinated APIs. Those results describe the tested cases and should not be treated as a guarantee that every defect will be detected.