# ai-architect-mcp-spec

Validates PRDs and specs against code graphs and requirements before code generation begins. It catches hallucinated APIs and unlinked acceptance criteria early.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed · ★ 6
- Author: cdeust
- License: MIT · policy=link-only · description only; the file stays with the author
- Source: https://github.com/cdeust/ai-architect-mcp-spec
- Updated: 2026-10-23
- HTML: https://affpapa.org/en/nevibecode/mcp/ai-architect-mcp-spec

## What it does

AI coding agents frequently hallucinate missing functions or invent performance targets during spec generation. This MCP server prevents those errors before any code gets written. It validates referenced symbols against a codebase graph and confirms that every acceptance criterion traces to a documented requirement. The verification rules run deterministically without calling a language model in the checking loop.

The project bundles 17 MCP tools across 20 pipeline stages. It operates as a stateless reducer where the server yields actions for the host environment to run. A panel of reasoning agents evaluates critical claims, logging outcomes to a local SQLite database for Bayesian reliability calibration. Consensus is monitored closely: an identical verdict across five or more claims triggers a suspicious distribution flag.

Local execution requires Node.js 20.x or 22.x and TypeScript 7.0+. The full generation workflow currently runs only in Claude Code, though other stdio hosts can execute the deterministic validator. Performance claims never receive an automatic pass without empirical benchmarks, forcing a status of `SPEC-COMPLETE` or `NEEDS-RUNTIME`.

## Install

### Claude Code (native platform)

**Install**
- The MCP registry has no launch command for it: see the repository README

## When to use it
- verify PRD symbols against the codebase graph
- trace acceptance criteria back to functional requirements
- block unmeasured latency and throughput claims
- calibrate judge agent reliability with Bayesian posteriors

## Example prompts
- Verify this PRD draft against the codebase graph
- Run the step-by-step PRD generation pipeline with acceptance criteria checks
- Check if the judge panel consensus on these requirements looks suspicious

## Similar
- [REA](https://affpapa.org/en/nevibecode/mcp/rea.md) — MCP server · ★ 392262
- [GitHub](https://affpapa.org/en/nevibecode/mcp/github.md) — MCP server · ★ 33989
- [Netdata](https://affpapa.org/en/nevibecode/mcp/netdata.md) — MCP server · ★ 81163
- [browser-use](https://affpapa.org/en/nevibecode/mcp/browser-use.md) — MCP server · ★ 119231
- [Codebase Memory](https://affpapa.org/en/nevibecode/mcp/codebase-memory.md) — MCP server · ★ 47577
- [mcp-server](https://affpapa.org/en/nevibecode/mcp/mcp-server.md) — MCP server · ★ 43902

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md