iphone-safari-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| IPHONE_SAFARI_DRIVER_URL | No | URL of an existing safaridriver instance to attach to, for debugging. If not set, the server starts its own safaridriver. | |
| IPHONE_SAFARI_DEVICE_NAME | No | Name of the iPhone device to use. | |
| IPHONE_SAFARI_DEVICE_TYPE | No | Type of the iPhone device to use (e.g. iPhone, iPad). | |
| IPHONE_SAFARI_DEVICE_UDID | No | UDID of the iPhone device to use. | |
| IPHONE_SAFARI_DISCOVERY_TIMEOUT | No | Number of seconds to retry session creation for safaridriver device discovery. Defaults to 45. | 45 |
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": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| session_infoA | Report what device/browser this session is actually attached to. Use this first to confirm you are driving the real iPhone and not Mac Safari. |
| navigateA | Open a URL in Safari on the iPhone and wait for load. |
| screenshotC | Screenshot the current iPhone Safari viewport. |
| snapshotA | Compact structured view of the page: URL, viewport, scroll position, and every visible interactive element with a CSS selector, label and on-screen rect. Prefer this over page_source() — it is the cheap way to decide what to click. |
| page_sourceA | Raw HTML of the current page, truncated to max_chars. |
| clickC | Click/tap an element. |
| type_textA | Type into an input on the phone, optionally clearing it first or submitting the form. |
| tapA | Send a real single-finger touch at viewport coordinates (CSS pixels). Use when there is no sensible selector — e.g. a canvas or a custom overlay. |
| scrollA | Scroll the page by a pixel delta. Positive dy scrolls down. |
| evaluate_javascriptA | Run JavaScript in the page and return the result as a string. Wrap in |
| console_logsA | Console messages and uncaught errors captured since the last navigation. The capture hook is installed by navigate(); if a page navigated itself, call install_console_capture() to re-arm it. |
| install_console_captureA | (Re)install the console/error capture hook on the current page. |
| network_timingsB | Resource timings from the Performance API — URL, type, duration, transfer size. A zero transferSize on a non-cached entry usually means the request failed. |
| go_backA | Navigate back in Safari's history. |
| reload_pageA | Reload the current page and re-arm console capture. |
| close_sessionA | End the automation session on the phone. A later call reconnects automatically. |
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 16 tools
Every tool has a clearly distinct purpose: navigate vs. reload_page vs. go_back are separate navigation actions; click vs. tap distinguish element-based from coordinate-based interaction; screenshot, snapshot, and page_source each offer a different view of the page. There is no meaningful overlap that would cause an agent to misselect.
Most tool names follow a verb_noun pattern (navigate, reload_page, evaluate_javascript, install_console_capture), with a few single-verb commands (click, tap, scroll) and noun-like names (screenshot, snapshot, page_source) that act as commands. All names are lowercase snake_case, which keeps the set predictable and readable despite minor pattern deviations.
With 16 tools, this server sits just above the ideal 3-15 range, but each tool covers a distinct aspect of mobile browser automation—navigation, interaction, inspection, and session control. The count is justified by the breadth of features needed for iPhone Safari and does not feel bloated or redundant.
The tool set covers the core lifecycle of a browsing session: navigate, go_back, reload, click/tap/scroll/type, extract page data (snapshot, page_source, screenshot), execute JS, and capture console/network diagnostics. Minor gaps exist (e.g., no explicit wait-for-element or cookie management), but they are workable and the server is functionally complete for typical automation tasks.