Gemini Bridge
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| AGY_BIN | No | The agy executable. | %LOCALAPPDATA%\agy\bin\agy.exe |
| IMAGE_DIR | No | Where generated images land. | %USERPROFILE%\Pictures\gemini-bridge |
| MODEL_PRO | No | Only when the caller asks for Pro. | gemini-3.1-pro-high |
| VAULT_DIR | No | Where deep research is saved as notes. Unset: research is answered inline and nothing is saved. | |
| AGY_ACCOUNT | No | The Google account agy must be signed in with. A run agy logs under another account is discarded (wrong_account). | |
| AGY_APP_DIR | No | agy's data folder: settings.json, transcripts, artifacts. | %USERPROFILE%\.gemini\antigravity-cli |
| BRIDGE_HOME | No | Jobs, scratch workspaces and runs.jsonl. | %LOCALAPPDATA%\gemini-bridge |
| MODEL_FLASH | No | Quick research, read, image, ask. | gemini-3.8-flash-medium |
| BLOCKED_DIRS | No | ';'-separated directories that never reach Gemini, as file inputs or as paths pasted into a prompt. | |
| CONTENT_RULE | No | The never-send sentence in every tool description, which is what Claude reads before delegating. | Personal content only — never work code, client or customer data. |
| SYNC_BUDGET_MS | No | How long a call waits before it hands back a job id. | 150000 |
| RESEARCH_SUBDIR | No | Folder inside VAULT_DIR for saved research. | 50-Research/gemini-researches |
| MODEL_FLASH_DEEP | No | Deep research. | gemini-3.8-flash-high |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| gemini_researchA | Web research on your Google AI Pro quota (Gemini + Google Search). Use for sweeps across many sources, current information, or to save Claude context. Long runs return a job id; check it with gemini_job. Personal content only — never work code, client or customer data. Gemini cannot see this conversation, so put all context in |
| gemini_askA | Ask Gemini directly (no tools, no web) on your Google AI Pro quota: second opinions, rewrites, quick reasoning, critique of a plan or a draft. Gemini cannot see this conversation, so include the full context in |
| gemini_readA | Have Gemini read one local file or one web page and answer a question about it, on your Google AI Pro quota. Handles PDFs, audio (transcription), video (what is shown and said), images and web pages, natively and multimodally — use it to save Claude context on large or media files. Give exactly one of |
| gemini_imageA | Generate one image with Gemini on your Google AI Pro quota and save it in %USERPROFILE%\Pictures\gemini-bridge as -. (usually 1024×1024 JPEG). Returns the file path. Personal content only — never work code, client or customer data. |
| gemini_jobA | Check a Gemini bridge job: status, then the result (or saved file path) once it has finished, or the error. Works for jobs started from either Claude Code or Claude Desktop/Cowork. Poll every minute or two while it is running. |
| gemini_quotaA | Show your remaining Gemini quota on Google AI Pro (weekly and 5-hour windows, from agy /usage, which spends no quota) plus today's bridge runs and tokens. Check it before a batch of deep research; a used-up weekly limit can lock Gemini out for days. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 6 tools
Each tool has a clearly distinct purpose: direct no-web asking, web research, reading a file/page, image generation, job polling, and quota inspection. The descriptions explicitly separate ask from research and read, so an agent should select the right tool without ambiguity.
All tools follow a consistent gemini_<capability> pattern with short, predictable suffixes. The mix of verbs and nouns (ask, research, read, image, job, quota) is still coherent because it is capability-based rather than an inconsistent naming convention.
Six tools is well-scoped for a Gemini bridge: core interaction modes, media handling, async job polling, and quota visibility. No tool feels redundant or superfluous.
The surface covers asking, web research, multimodal file/page reading, image generation, job status, and quota checks, which are the core bridge workflows. Minor gaps exist around job lifecycle management, such as listing or cancelling jobs, but agents can work around these.