What did Hlido score agy-bridge?

agy-bridge scored 66/100 (FADING) in Hlido's independent, hands-on review.

Does any vendor pay Hlido for placement?

No. Hlido takes no money from the agents it rates — scoring weights stay private and the evidence behind every verdict is public.

agy-bridge

Name: agy-bridge review
Item: agy-bridge
Rating: 66
Author: Hlido Editor

Coding · tested 2026-08-18 · re-test due 2026-11-16 · by the Hlido desk, not the vendor

In short: A small, unusually well-engineered MCP bridge that lets Claude Code offload token-heavy grunt work to Google's Antigravity CLI — high craft, honest about its own failure modes, but a niche tool for a specific two-CLI setup.

Visit agy-bridge → scorecard.json

Quick answer

agy-bridge scores 66/100 (FADING) on Hlido’s independent, hands-on test (reviewed 2026-08-18). FADING (66) — a deliberately narrow score for a deliberately narrow tool.

Hlido Editor · 2026-08-18

agy-bridge does one narrow thing and does it with more engineering discipline than most products ten times its size. The problem is concrete: Claude Code burns its context window reading raw material — a 3,000-line log, a year of git history — when what you want is the synthesized answer. agy-bridge is an MCP server (on npm, zero-install via npx) that lets Claude Code delegate that heavy reading to Google's Antigravity CLI (agy), which has its own model quota and local file access; only the summary comes back. What lifts it above a weekend hack is the thinking on display in the author's writeup. Rather than one generic 'ask Gemini' tool it exposes six purpose-built tools (analyze_files, deep_search, web_lookup, adversarial_review, follow_up, delegate), because tool descriptions are how the calling model decides when to delegate — a genuinely non-obvious insight. It stays stateless while offering session continuity by reading agy's own conversation cache. Its model routing degrades gracefully because preview-CLI model names change weekly. Most tellingly, v0.2.0 added AGY_ON_FAILURE=strict after the author noticed the worst failure was 'success theater' — a silent fallback where Claude quietly did the heavy work itself, recreating the exact problem the tool exists to solve — and concluded that instructing the calling model what to do on failure is part of an MCP tool's contract. Output is capped to protect the context window; 35 unit tests run without agy installed. The honest limits are inherent to what it is: it requires both Claude Code AND the Antigravity CLI installed, it is a solo project at v0.2, single-purpose, and the only surface we could review is the author's own blog post — no independent adoption or reliability signal. As a piece of craft and a model of how to write an MCP tool, it punches far above its scope; as a product, it is a sharp niche utility.

Why FADING

FADING (66) — a deliberately narrow score for a deliberately narrow tool. The engineering craft is high (six purpose-built tools with model-aware descriptions, stateless session continuity, graceful model routing, a strict fail-mode that closes the 'silent fallback' hole, 35 CLI-free unit tests) and the design reasoning is exemplary. It sits below STEADY because it is single-purpose and setup-dependent (needs both Claude Code and Google's Antigravity CLI), a solo project at v0.2, and the only reviewable surface is the author's blog post — no independent adoption, reliability or usage evidence.

What it does well

Offloads token-heavy reading (large logs, git archaeology) from Claude Code to agy, returning only the synthesized answer
Exposes six purpose-built tools whose descriptions give the calling model concrete triggers to delegate — not one generic tool
Stays stateless while offering session continuity by reading agy's own conversation cache and returning a follow-up ID
Routes each tool through a graceful model-preference chain, so churning preview-CLI model names don't break it
Fails loudly by design: AGY_ON_FAILURE=strict stops Claude from silently doing the heavy work itself
Caps output to protect the very context window it exists to save; 35 unit tests run with no agy install or network

What it fails at

Requires both Claude Code and Google's Antigravity CLI (agy) installed — a narrow two-tool prerequisite
Single-purpose delegation bridge; not a general-purpose agent capability
Solo project at v0.2 with no independent adoption, reliability or usage evidence on the surface
The only reviewable surface is the author's own blog post — no product site, docs portal or third-party validation
Value depends on agy's Gemini quota and behaviour, an external preview CLI outside the bridge's control

Best for

Developers already running Claude Code alongside Google's Antigravity CLI who want to conserve Claude's context window
Workflows with recurring token-heavy tasks — large-file analysis, repo-wide git archaeology, adversarial code review
Engineers who value a delegation tool that fails loudly rather than silently reverting to expensive local work
Anyone studying how to design MCP tools well — the writeup is a strong reference

Not recommended for

Users who don't run Google's Antigravity CLI (the bridge has nothing to delegate to)
Teams needing a supported, multi-contributor product with reliability guarantees
General agent tasks beyond offloading heavy reading/analysis
Buyers who require documentation and adoption evidence beyond an author blog post

Related agents

Agent relevance

CLI MCP Behavioral-testable

Agentic-Commerce Readiness 50/100 · INTEGRABLE

Independent readiness for agent delegation & transaction. How it’s scored · check live

First-class MCP for Claude Code: add agy-bridge to mcpServers (command npx, args -y agy-bridge) and Claude Code gains six delegation tools. It is agent-facing by construction — a bridge one agent (Claude Code) calls to route heavy work to another CLI (agy/Gemini). Requires the Antigravity CLI installed locally to function.

Agent-friendly score: 7/10

Evidence

MCP server on npm (npx -y agy-bridge) delegating heavy work from Claude Code to agy/Gemini — source (2026-08-18) verified
Six purpose-built tools with per-tool model routing chains — source (2026-08-18) verified
AGY_ON_FAILURE=strict prevents silent local fallback; output capped; 35 unit tests — source (2026-08-18) verified
Independent adoption or reliability evidence beyond the author's blog — source (2026-08-18)

scorecard.json · registry · methodology

More: compare agents · best of · developer tools · incident registry

Verdict by Hlido Editor, our automated editorial system · Method: public-surface-tier-1+editorial-narrative-v2 · Methodology version 2026.05 · Next review due 2026-11-16

How this page was produced. The scores, claim verdicts and evidence come from automated hands-on testing of the product’s public surface. The written analysis is drafted by an AI system, and pages publish without a person reviewing each one. Hlido publishes this record and answers for it — tell us if anything here is wrong and we will correct it.

Embed this trust badge

Live, always-current independent score — free to embed on your site or README. No vendor pays for placement.

Markdown

[![Hlido trust score](https://hlido.eu/badge/sshahzaiib-agy-bridge.svg)](https://hlido.eu/check/?agent=sshahzaiib-agy-bridge)

HTML

<a href="https://hlido.eu/check/?agent=sshahzaiib-agy-bridge"><img src="https://hlido.eu/badge/sshahzaiib-agy-bridge.svg" alt="Hlido trust score"></a>