📡 MCP Radar

kanetran29/uigraph

Your app generates its own verified UI transition graph — screens as nodes, guarded (event, effect) edges. Proof-gated verification, trust tiers, model-free MCP for AI agents.

⭐ 3 stars
⑂ 0 forks
TypeScript
momentum ▲ 6.0
created 2026-07-12
on radar since 2026-07-15
ai-agentsangularcode-graphe2e-testingmcpmodel-context-protocolnextjsplaywrightreactstatic-analysistest-generationtypescriptui-graphvue
View on GitHub ↗Homepage ↗

About uigraph

Your app generates its own verified behavioral map. React · Vue · Angular · Next.js — screens as nodes, (event, guard, effect) transitions as edges.

Tests, docs, agents, and impact analysis are all views over one graph — and every edge says how much you can trust it.

Your agent re-derives your app's navigation from source on every session — and a graph it guesses today can disagree with the one it guesses tomorrow. UI-graph computes the map once per commit, deterministically, then makes the LLM's additions earn their way in with runtime proof. The moat is not the word "map". It's the word verified.

[!NOTE] The core is model-free: no API key, no bundled LLM. Your coding agent (Claude Code, Cursor, Copilot) connects over MCP and brings its own model — the graph works offline and in LLM-restricted CI.

--- --- Deterministic extraction ts-morph static analysis of react-router v5/v6/data-router, vue-router, Angular Routes, and Next.js file routes — every edge carries a file:line witness Proof-gated verification reportobservation(confirmed) requires evidence (URL change, asserted dialog, screenshot on disk) + provenance, or it is rejected. A hallucinated confirmation cannot enter the graph Trust tiers on every edge witnessed proven asserted llm-verified proposed unknown — projected on read, so an agent always knows how far to lean The Tier-2/3 loop The LLM proposes greedily (propose, quarantined), the runner drives a real browser (uigraph verify), and only the observation — never the guess — mints an edge Honest coverage runtime-verified ⊆ verified ⊆ accounted are never confla

From the project README.

Maintaining this server?

Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:

[![On MCP Radar](https://mcp.liqiwa.com/badge.svg)](https://mcp.liqiwa.com/s/kanetran29--uigraph.html)

Related MCP servers

coffeecoproject/layermap

LayerMap MCP: a layered map of your codebase for coding agents: callers, callees and the impact of a change, for TypeScript, JavaScript, Go, Python and Java. MCP server, CLI, and plugins for Claude Co

⭐ 5TypeScript
cyberchronos00/code-graph

A deterministic code graph for codebases: every caller, route, page and table a change touches, with file:line evidence. CLI, MCP server and web view for AI agents and reviewers.

⭐ 4Python
Quality-Max/9lives-runner

Standalone Go runner and @9l/playwright SDK for bounded Playwright execution with validated receipts.

⭐ 4Go
herry2059/project-os-for-codex

Open-source control plane for Codex projects: Git-backed context, visible agent progress, scoped MCP access, resumable work, and safe handoffs.

⭐ 104TypeScript
roshanarnav25-sloth/google-flow-mcp

MCP server that drives Google Flow (Veo / Nano Banana) through its internal tRPC API, with a hard credit-cost gate

⭐ 8TypeScript
MarlonJD/web-debug-mcp

Evidence-first local MCP debugger for React, Vite, Next.js, Chromium, and Safari: reproduce, inspect, replay, and verify web bugs in one bounded workflow.

⭐ 6TypeScript

📬 Get the weekly radar in your inbox

The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.