Skip to main content
Glama
DustinTrap

kvm-pilot

by DustinTrap
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Remote

mouse

Destructive

Moves the host cursor to a target coordinate and optionally clicks a button. Click actions verify a recent snapshot reference to block stale or foreign input.

Instructions

Move the mouse (and optionally click) on the host. DESTRUCTIVE (HID input).

A click must carry the observed_frame_ref it was planned against (from a prior snapshot); it is refused — re-snapshot and retry — if the host rebooted/swapped media since, the observation is older than KVM_PILOT_MCP_FRAME_MAX_AGE (60s default, #141), or the ref wasn't issued by this server. Move-only (button omitted) needs no ref. coord_space: percent (0.0-1.0, default — survives resolution changes), pixel, or raw kvmd. Gated by KVM_PILOT_MCP_ALLOW_HID

  • per-invocation approval.

Input Schema

TableJSON Schema
NameRequiredDescriptionDefault
xYesHorizontal target in coord_space units (percent: 0.0 left to 1.0 right).
yYesVertical target in coord_space units (percent: 0.0 top to 1.0 bottom).
buttonNoButton to click after moving. Omit for a move only (no frame ref needed).
confirmNoPer-invocation approval when the client cannot prompt a human: set true to approve this exact call under the operator's standing policy. Leave false when the client supports elicitation (a human is asked). Never pre-set it speculatively; a denial returns outcome=not_confirmed.
profileNoProfile name from ~/.config/kvm-pilot/config.toml selecting which KVM/BMC (and its managed host) this call targets. Omit to use the server's default (KVM_PILOT_PROFILE).
coord_spaceNopercent = fraction of the screen (survives resolution changes); pixel = capture pixels; raw = kvmd units.percent
observed_frame_refNoframe_ref from the snapshot the click was planned on. Required for a click; a stale or foreign ref is refused.

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed7 schema fields changedv1.0.0
    • addedInput schema / properties / button / description
      Added value: +"Button to click after moving. Omit for a move only (no frame ref needed)."
    • addedInput schema / properties / confirm / description
      Added value: +"Per-invocation approval when the client cannot prompt a human: set true to approve this exact call under the operator's standing policy. Leave false when the client supports elicitation (a human is asked). Never pre-set it speculatively; a denial returns outcome=not_confirmed."
    • addedInput schema / properties / coord_space / description
      Added value: +"percent = fraction of the screen (survives resolution changes); pixel = capture pixels; raw = kvmd units."
    • addedInput schema / properties / observed_frame_ref / description
      Added value: +"frame_ref from the snapshot the click was planned on. Required for a click; a stale or foreign ref is refused."
    • addedInput schema / properties / profile / description
      Added value: +"Profile name from ~/.config/kvm-pilot/config.toml selecting which KVM/BMC (and its managed host) this call targets. Omit to use the server's default (KVM_PILOT_PROFILE)."
    • addedInput schema / properties / x / description
      Added value: +"Horizontal target in coord_space units (percent: 0.0 left to 1.0 right)."
    • addedInput schema / properties / y / description
      Added value: +"Vertical target in coord_space units (percent: 0.0 top to 1.0 bottom)."
  2. First observedv0.1.0

TDQS

A4.5/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Far beyond the annotations (which only give destructiveHint=true), it discloses the exact refusal conditions (host reboot/media swap, ref older than KVM_PILOT_MCP_FRAME_MAX_AGE of 60s, foreign ref), the env gating (KVM_PILOT_MCP_ALLOW_HID) plus per-invocation approval, and the not_confirmed outcome. This is unusually rich behavioral disclosure.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Purpose is front-loaded in the first sentence, followed by the click-vs-move constraint and coordinate semantics. It is dense with backtick-heavy detail but every sentence carries actionable information; only minor tightening is possible.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a destructive, approval-gated HID tool with no output schema, it covers preconditions, gating, and one outcome value (not_confirmed). A little more on the returned result shape would fully close the loop, but the agent has what it needs to call correctly.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters4/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100% so the baseline is 3, but the description adds rationale and failure semantics the schema lacks: coord_space 'survives resolution changes', why the frame ref is needed, and the confirm/approval interaction. It reinforces rather than merely repeats the schema.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The first sentence states a specific verb and resource ('Move the mouse (and optionally click) on the host') and immediately flags the operation class ('DESTRUCTIVE (HID input)'). This is clearly distinguishable from siblings like calibrate_mouse or press_key.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

It gives a clear conditional rule: a click requires an observed_frame_ref from a prior snapshot, while a move-only call needs none, and it directs the agent to re-snapshot and retry on refusal. It stops short of explicitly naming a sibling to use instead, but the snapshot-first workflow is unambiguous.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Deploy Server

Other Tools