agy-bridge

An MCP bridge that lets Claude Code delegate heavy tasks to the Antigravity CLI (agy) — saving Claude's context window and tokens for what matters.

Claude sends a task → the bridge routes it to the best available model via agy → only the answer comes back. Large files, deep git searches, and web lookups never touch Claude's context.

User → Claude Code → agy-bridge (MCP) → agy CLI → Gemini / Claude / GPT-OSS
                   ←                  ←          ←

Why this over claude-to-agy?

Related MCP server: Herald

Requirements

• Node.js 18+

• Antigravity CLI (agy) installed and authenticated

• Claude Code

Install

# 1. Register the MCP server (user scope = all projects)
claude mcp add -s user agy-bridge npx -- -y agy-bridge

# 2. Add delegation rules to your project (or ~/.claude/CLAUDE.md for global)
curl -o CLAUDE.md https://raw.githubusercontent.com/sshahzaiib/agy-bridge/main/CLAUDE.md

Tools

All tools accept optional cwd (project root) and model (exact name from agy models; validated, with available models listed on mismatch).

Every response ends with a footer:

---
[agy-bridge] model: Gemini 3.5 Flash (High) | session: 1f0c…-d4 (use follow_up to continue)

Model routing

On first use the bridge runs agy models (cached for the process lifetime) and picks the first available model in the tool's preference chain. If none is available it falls back to AGY_DEFAULT_MODEL, and finally to agy's own default. agy silently ignores unknown --model values, so the bridge validates names up front instead of letting requests land on the wrong model.

Quota-aware failover

agy never surfaces quota exhaustion in print mode — it silently retries the 429 until its print-timeout, then exits 0 with empty output, which used to look like an indefinite hang. The bridge now watches each run's log file (via --log-file) and on RESOURCE_EXHAUSTED (code 429):

1. kills the agy process group immediately (no waiting out the timeout),

2. parses the reset time ("Resets in 4h24m") into an in-process cooldown registry,

3. retries the same prompt on the next model in the tool's chain,

4. skips cooled-down models on all subsequent calls until their quota resets.

Failovers are annotated in the response footer (failover: <model>: quota exhausted (resets in 4h24m)). Only when every candidate is exhausted does the call fail — in seconds, with reset times listed — instead of hanging.

Timeouts and cancellation

Each tool has its own default timeout sized to its job: web_lookup 120s, deep_search 180s, analyze_files / adversarial_review / follow_up 300s, delegate 600s. Setting AGY_TIMEOUT explicitly overrides all of them. The kill path escalates SIGTERM → SIGKILL across the whole process group, and the deadline fires even if agy's helper processes hold the output pipes open. Cancelling the tool call from the MCP client (e.g. pressing Esc in Claude Code) also kills the agy run instead of orphaning it.

Configuration

All optional, via environment variables:

Failure behavior

The bridge always fails loudly: agy errors surface as MCP tool errors with agy's actual stderr, and degraded model routing is annotated in the response footer. By default the calling agent (Claude) will typically do the work itself after a failure — visible in the transcript, but easy to stop noticing in a long session. Set AGY_ON_FAILURE=strict to append an explicit "do NOT perform this work yourself — report the failure to the user" instruction to every delegation error, so you keep control over when token savings are silently lost.

Development

npm install
npm test           # vitest unit tests (exec mocked — no agy needed)
npm run typecheck
npm run build      # tsup → dist/index.js

Contributors

Contributions are welcome — open an issue or PR.

Star History

License

MIT