---
title: "MCP tools"
summary: "The 14 read-only tools `driga mcp` serves to your agents and the 8 tools only Driga's own Chat receives, with their inputs."
description: "Every tool driga's MCP server exposes, with its inputs, generated from the server's tools/list answer."
generated: true
read_when:
  - "You are connecting an agent to driga over MCP and want to know what it can call."
  - "You need a tool's exact input fields, defaults or limits."
---

`driga mcp` serves read-only tools over stdio. Every response is the same
schema-2 view envelope as the CLI's `--json` output, and reading never
acknowledges inbox items. Without a running collector the tools answer from a
labelled one-shot snapshot with no history. Registration is covered in
[Connect an agent over MCP](/guides/mcp).

Every read tool shares this contract:

> Use this to inspect collected state without changing it; safe to retry. Optional scope fields select machine (default all) and project where supported; for example \{"limit":100\}. Returns a schema-2 envelope with data, UTC observed_at, source freshness and coverage gaps; mode: "snapshot" means no history. limit bounds total array entries, including nested collections (default 100, maximum 500). truncated=true means entries were omitted: narrow the scope or increase limit. Reading never acknowledges inbox items.

## Read tools

Served by `driga mcp`. All 14 are annotated read-only.

| Tool | Read-only |
|---|---|
| [`get_activity`](#get_activity) | yes |
| [`get_agent`](#get_agent) | yes |
| [`get_flow`](#get_flow) | yes |
| [`get_machine`](#get_machine) | yes |
| [`get_summary`](#get_summary) | yes |
| [`inspect_project`](#inspect_project) | yes |
| [`inspect_setup`](#inspect_setup) | yes |
| [`list_agents`](#list_agents) | yes |
| [`list_cleanup_candidates`](#list_cleanup_candidates) | yes |
| [`list_conflicts`](#list_conflicts) | yes |
| [`list_inbox`](#list_inbox) | yes |
| [`list_machines`](#list_machines) | yes |
| [`list_orphans`](#list_orphans) | yes |
| [`list_projects`](#list_projects) | yes |

### get_activity

Read the last sixty minutes of machine samples and event counts.

| Input | Type | Required | Description |
|---|---|---|---|
| `limit` | integer |  | Maximum total array entries, including nested collections; 1–500, default 100. |
| `machine` | string |  | Machine name, ID, device-ID prefix, local, or all (default all). |

### get_agent

Read one session's detail and recent evidence; use required id from list_agents, e.g. id=1.

| Input | Type | Required | Description |
|---|---|---|---|
| `id` | integer | yes | Internal session ID shown by driga agents. Minimum 1. |
| `limit` | integer |  | Maximum total array entries, including nested collections; 1–500, default 100. |
| `machine` | string |  | Machine name, ID, device-ID prefix, local, or all. |

### get_flow

Inspect one session family map: per-node stored mixer, events, inbox wait spans, observations, retention, gaps and machine-wide waiting attention. Optional id from list_agents selects its lead; omit for the newest live interactive session. Missing telemetry measurements are omitted; usage covers retained api_request events only.

| Input | Type | Required | Description |
|---|---|---|---|
| `id` | integer |  | Minimum 1. |
| `limit` | integer |  | Maximum total array entries, including nested collections; 1–500, default 100. |
| `machine` | string |  |  |
| `minutes` | integer |  | Time window: 15, 60 or 240 minutes; default 60. |
| `project` | string |  |  |

### get_machine

Read agents' share of machine memory and CPU, cost by session, agent processes without a session, likely agent work, other apps, MCP server roll-ups and pile-ups. Defaults to bounded detail and 24h trends; full includes all retained evidence.

| Input | Type | Required | Description |
|---|---|---|---|
| `full` | boolean |  |  |
| `limit` | integer |  | Maximum total array entries, including nested collections; 1–500, default 100. |
| `machine` | string |  |  |

### get_summary

Read agent counts, inbox counts, and machine health.

| Input | Type | Required | Description |
|---|---|---|---|
| `limit` | integer |  | Maximum total array entries, including nested collections; 1–500, default 100. |
| `machine` | string |  | Machine name, ID, device-ID prefix, local, or all (default all). |

### inspect_project

Read a project's sourced status, worktrees, and handoffs; required project is a name or path, e.g. project=api.

| Input | Type | Required | Description |
|---|---|---|---|
| `limit` | integer |  | Maximum total array entries, including nested collections; 1–500, default 100. |
| `project` | string | yes | Project name or path. Must not be empty. |

### inspect_setup

Read setup and collection diagnostics.

| Input | Type | Required | Description |
|---|---|---|---|
| `limit` | integer |  | Maximum total array entries, including nested collections; 1–500, default 100. |

### list_agents

Read agent sessions and their families.

| Input | Type | Required | Description |
|---|---|---|---|
| `all` | boolean |  | Include sessions ended in the last 24 hours. |
| `limit` | integer |  | Maximum total array entries, including nested collections; 1–500, default 100. |
| `machine` | string |  | Machine name, ID, device-ID prefix, local, or all. |
| `project` | string |  | Project name, path, or dot for the local working directory. |

### list_cleanup_candidates

Inspect local disk cleanup candidates read-only. Optional kind is worktree, scratch, or history (omit for all); sizes=true opts into bounded directory sizing. This tool never removes anything.

| Input | Type | Required | Description |
|---|---|---|---|
| `kind` | string |  | One of `worktree`, `scratch`, `history`. |
| `limit` | integer |  | Maximum total array entries, including nested collections; 1–500, default 100. |
| `sizes` | boolean |  |  |

### list_conflicts

Read current and recent file overlaps.

| Input | Type | Required | Description |
|---|---|---|---|
| `limit` | integer |  | Maximum total array entries, including nested collections; 1–500, default 100. |
| `machine` | string |  | Machine name, ID, device-ID prefix, local, or all. |
| `project` | string |  | Project name, path, or dot for the local working directory. |

### list_inbox

Read attention and completion events.

| Input | Type | Required | Description |
|---|---|---|---|
| `all` | boolean |  | Include acknowledged events. |
| `limit` | integer |  | Maximum total array entries, including nested collections; 1–500, default 100. |
| `machine` | string |  | Machine name, ID, device-ID prefix, local, or all. |
| `project` | string |  | Project name, path, or dot for the local working directory. |
| `since` | string |  | Non-negative duration or yesterday. |

### list_machines

Read cached fleet connection health, even when collection has stopped.

| Input | Type | Required | Description |
|---|---|---|---|
| `limit` | integer |  | Maximum total array entries, including nested collections; 1–500, default 100. |

### list_orphans

Read orphan process candidates without signaling processes.

| Input | Type | Required | Description |
|---|---|---|---|
| `limit` | integer |  | Maximum total array entries, including nested collections; 1–500, default 100. |
| `machine` | string |  | Machine name, ID, device-ID prefix, local, or all (default all). |

### list_projects

Read project activity and risk flags.

| Input | Type | Required | Description |
|---|---|---|---|
| `active` | boolean |  | Only active projects. |
| `at_risk` | boolean |  | Only projects with risk flags. |
| `limit` | integer |  | Maximum total array entries, including nested collections; 1–500, default 100. |
| `machine` | string |  | Machine name, ID, device-ID prefix, local, or all. |

## Chat tools

Offered only to Driga's own Chat turn through a private MCP child, never by
`driga mcp`. Action tools run through Chat's approval cards and trust modes;
see [Chat and actions](/guides/chat).

| Tool | Read-only |
|---|---|
| [`ack_inbox`](#ack_inbox) | no |
| [`answer_prompt`](#answer_prompt) | no |
| [`approve`](#approve) | no |
| [`cleanup`](#cleanup) | no |
| [`jump`](#jump) | no |
| [`kill`](#kill) | no |
| [`list_prompts`](#list_prompts) | yes |
| [`summarize`](#summarize) | no |

### ack_inbox

Acknowledge a local inbox id or all local items.

| Input | Type | Required | Description |
|---|---|---|---|
| `all` | boolean |  |  |
| `id` | string |  |  |

### answer_prompt

Answer one pending local Claude permission prompt. Always asks in Auto.

| Input | Type | Required | Description |
|---|---|---|---|
| `allow` | boolean | yes |  |
| `id` | string | yes |  |

### approve

Request driga permission for a tool. Cannot decide cards or change trust mode.

| Input | Type | Required | Description |
|---|---|---|---|
| `input` | object | yes |  |
| `tool_name` | string | yes |  |
| `tool_use_id` | string | yes |  |

### cleanup

Clean one local candidate after re-verification.

| Input | Type | Required | Description |
|---|---|---|---|
| `id` | string |  |  |

### jump

Jump to a local session terminal. Never types into the terminal.

| Input | Type | Required | Description |
|---|---|---|---|
| `session_id` | integer |  |  |

### kill

Kill a local agent session or orphan id after identity and idle verification. Force also escalates survivors.

| Input | Type | Required | Description |
|---|---|---|---|
| `force` | boolean |  |  |
| `id` | string |  |  |
| `session_id` | integer |  |  |

### list_prompts

Read pending local Claude permission prompts. Inputs are visible only under snippets capture.

No inputs.

### summarize

Refresh one local project summary.

| Input | Type | Required | Description |
|---|---|---|---|
| `project` | string |  |  |