agy-agent-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| AGY_BIN | No | Path to the Antigravity CLI binary. | agy |
| AGY_MCP_ROOT | No | Directory jail for task cwd and edit_scope paths. | |
| AGY_MCP_STATE_DIR | No | Directory for per-run task_<id>.ndjson and task_<id>.stream files. | /tmp/agy-agent-mcp/tasks |
| AGY_MCP_RETENTION_DAYS | No | Days to retain .ndjson and .stream files on startup sweep (0 disables). | 7 |
| AGY_MCP_IDLE_TIMEOUT_MS | No | Per-run idle timeout in ms (15 min, resets on any child output; 0 disables). | 900000 |
| AGY_MCP_PRINT_TIMEOUT_MS | No | Opt-in hard wall-clock timeout per run in ms (--print-timeout + runner kill). | 0 |
| AGY_MCP_DENY_AUTO_APPROVE | No | Set to 1 or true to ignore auto_approve (--dangerously-skip-permissions). |
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 |
|---|---|
| general_taskA | Delegate a general-purpose coding task to the Antigravity (agy) CLI and block until the headless run exits (synchronous; the tool call IS the wait). Defaults: mode accept-edits, effort medium. The result carries the agent's final response, the resumable conversation_id, token usage, denied_actions, and the git-based files_touched/scope_violations audit when cwd is a git repo. Headless runs auto-DENY tool requests that would prompt instead of hanging (surfaced in denied_actions); permission to allow them is granted by the user via permissions.allow rules in their own Antigravity settings.json (e.g. command()), which this server never edits. File edits inside the task cwd never prompt; set edit_scope so out-of-scope changes are flagged in the audit. auto_approve is the expert full-auto switch (--dangerously-skip-permissions); pair it with sandbox when granting it. |
| explore_taskA | Delegate a fast read-only exploration and diagnosis task to the Antigravity (agy) CLI and block until the headless run exits (synchronous; the tool call IS the wait). Defaults: mode plan (read-only), effort low; the mode preset is fixed, so callers cannot weaken the read-only posture. Produces answers and summaries without modifying files. Returns the same report as general_task, including the git-based files_touched/scope_violations audit. Headless runs auto-DENY tool requests that would prompt instead of hanging (surfaced in denied_actions); permission to allow them is granted by the user via permissions.allow rules in their own Antigravity settings.json (e.g. command()), which this server never edits. auto_approve is the expert full-auto switch (--dangerously-skip-permissions); pair it with sandbox when granting it. |
| watch_taskA | Tails a task's live stream transcript. No args = latest run. Call it while a general_task/explore_task is in flight (this or any session) to watch the sub-agent work; the returned 'running' field plus 'modified' tell a live run from an abandoned one. |
| list_modelsA | List model IDs usable for the |
| list_agentsA | List agent profiles usable for the |
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 5 tools
Each tool has a clearly distinct purpose: general_task (writable coding delegation) vs explore_task (read-only plan mode) are separated by their mode posture, watch_task tails transcripts, and list_models/list_agents are unambiguous lookup helpers. No two tools compete for the same action.
All five names use snake_case with a consistent verb/noun shape (general_task, explore_task, watch_task, list_models, list_agents). The two list_* helpers and three *_task delegation tools form a predictable pattern.
Five tools is well-scoped for a delegation server: two delegation entry points, one monitoring tool, and two parameter-lookup helpers. Every tool earns its place with no redundant surface.
The core lifecycle of delegating, exploring, and observing tasks is covered, and resumable conversation_id implies continuation. However there is no explicit cancel/abort tool for an in-flight run and no task history/listing beyond 'latest run', which are minor but real gaps.