Flick
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| TYPESAFE_API_KEY | Yes | Your TypeSafe API key required for inference |
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 |
|---|---|
| computer_healthA | Check local setup without opening an app or making a paid model request. |
| computer_openB | Open a dedicated browser, connect to existing Chrome in a new task tab, or connect to a native macOS app. connection existing-chrome uses the running personal profile and Chrome’s user-approved remote-debugging flow. Supply allowedOrigins for additional sites needed by the task. Native OCR runs locally. |
| computer_sessionsA | List sessions owned by this local MCP process. |
| computer_appsA | List running macOS applications, or include installed apps for launchable targets. |
| computer_inspectA | Read visible controls and text. Use returned observation and element IDs for direct actions. Running task observations are available from computer_status. |
| computer_actA | Execute one precise action against a fresh observation, then return the resulting state. No selectors or executable code. For several steps prefer computer_run. |
| computer_runA | Start a bounded Jev automation task and immediately return its ID. Supply exact text in inputs and at least one observable until condition. All until conditions must pass. Jev chooses actions; code independently verifies completion. This tool operates the target app; only request actions the user authorized. |
| computer_executeA | Start one goal across a set of native apps and/or dedicated browsers. Targets are available apps, not ordered steps; Jev chooses the sequence and remembers exact observed text for reuse. If targets are omitted, discover installed macOS apps. Browser targets require a URL. Supply exact new text in inputs. All until conditions must pass; field_from_memory compares a destination to a remembered source named "App name / Field name". Returns task and session IDs immediately. Close the session when finished. |
| computer_continueA | Continue an unfinished stopped task with missing text values or additional guidance. Keeps the same apps, observed memory, and success conditions, then observes fresh state. Returns a new task ID linked to the previous attempt. |
| computer_workflowA | Start an ordered macOS workflow in one call. The local engine switches apps and runs Jev stages without returning to the assistant between clicks or apps. Supply goals, exact inputs, and observable conditions for each stage. inputsFrom can copy one observed field from an earlier zero-indexed stage. Holds desktop ownership until completion. Poll computer_status or stop with computer_cancel. Native apps stay open. |
| computer_statusA | Get a task’s status, independent verification, step log, and timings. waitMs can wait up to 20 seconds without busy polling. includeObservation returns the latest interface state. |
| computer_cancelB | Request cancellation. The current input operation may finish; no further action will be scheduled. Poll status for the terminal result. |
| computer_screenshotB | Return a PNG of the controlled browser tab or native window to the calling assistant. Native capture requires Screen Recording permission. |
| computer_copy_imageA | Copy a visible image from a fresh browser observation. Returns a PNG of the rendered image, its source URL, and optionally writes it to the macOS clipboard. This captures the image at its displayed resolution, not its original file resolution. |
| computer_closeA | Close an idle automation browser or release a native session. Native applications remain open. Cancel and wait for any running task first. |
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 15 tools
Multiple tools start automation tasks (computer_run, computer_execute, computer_workflow) with subtle differences that are difficult to distinguish, and computer_inspect/computer_status with includeObservation both return interface state. computer_screenshot and computer_copy_image also overlap in image capture, so boundaries between tools are not always clear.
All tools share the consistent 'computer_' prefix, but suffixes mix nouns (health, sessions, apps, status, workflow, screenshot) and verbs (open, inspect, act, run, execute, continue, cancel, close). This is readable and predictable at the prefix level, but it lacks a uniform verb_noun convention.
At 15 tools, the set sits at the upper edge of the typical well-scoped range, and the broad computer-automation domain justifies much of its size. However, the overlapping task-launching tools could be consolidated without losing coverage.
The core lifecycle is well covered: open, inspect, act, run, status, cancel, continue, and close, plus support utilities like screenshots, sessions, and app listing. Minor gaps exist, such as lack of fine-grained configuration or session management controls, but common workflows should not hit dead ends.