jev-ra
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| BU_CDP_URL | No | CDP URL of an existing Chrome to drive instead of launching a new one. | |
| JEV_RA_MODEL | No | Override the model name. | |
| JEV_RA_CHROME | No | Path to the Chrome binary to launch. | |
| JEV_RA_API_KEY | No | API key for TypeSafe (direct). Highest precedence; used if set. | |
| JEV_RA_ENDPOINT | No | Override the API endpoint. | |
| JEV_RA_VIEWPORT | No | Viewport size, e.g. 1280x900. | 1280x900 |
| JEV_RA_MAX_STEPS | No | Maximum number of steps per run. | 40 |
| JEV_RA_TIMEOUT_S | No | Timeout in seconds. | 120 |
| TYPESAFE_API_KEY | No | TypeSafe API key. Used if JEV_RA_API_KEY is not set. | |
| JEV_RA_SEARCH_URL | No | Search endpoint template; {query} is substituted. | |
| JEV_RA_TEXT_MODEL | No | Optional text helper model name. | |
| OPENROUTER_API_KEY | No | OpenRouter API key (starts with sk-or-). Used if neither JEV_RA_API_KEY nor TYPESAFE_API_KEY is set. | |
| JEV_RA_TEXT_API_KEY | No | Optional text helper API key. | |
| JEV_RA_MAX_DECISIONS | No | Maximum number of decisions per run. | 80 |
| JEV_RA_TEXT_BASE_URL | No | Optional text helper base URL. | |
| JEV_RA_BLOCK_RESOURCES | No | Set to '0' to stop blocking fonts and media. | 1 |
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 |
|---|---|
| browser_openA | Open a URL in the shared browser session. A named profile keeps its own cookies. |
| browser_runA | Pursue a whole goal on the current page. Supply values for anything that must be typed. Pass resume with the run id a needs_human escalation returned, once the person has cleared the check, to carry that run on instead of starting the goal again. |
| browser_actC | Take one decided step towards an instruction on the current page. |
| browser_searchA | Search the web, read the best results in parallel tabs, and rank them against the goal. |
| browser_observeA | List the observed controls and the visible text of the current page. |
| browser_extractB | Pull structured page data from the DOM: text, elements, links, tables or main. |
| browser_clickA | Click one observed element by its ref. Pass the page_key from browser_observe to refuse a stale ref. |
| browser_typeA | Type text into one observed field by its ref, refusing a page that moved since that observation. |
| browser_selectA | Select an observed dropdown option by its value or label, refusing a stale ref. |
| browser_scrollB | Scroll the page one viewport step up or down. |
| browser_pressB | Press Enter, Escape or Tab. |
| browser_waitA | Wait a moment and observe again. |
| browser_screenshotA | Capture the current viewport as a JPEG. |
| browser_closeA | Close the browser session held by this server. |
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 14 tools
Most tools have clear boundaries: open/search/observe/extract/click/type/select/scroll/press/wait/screenshot/close are distinct actions. The only mild overlap is between browser_observe and browser_extract, but their descriptions separate visible/control inspection from structured data extraction.
Every tool follows the same browser_<verb> pattern with lowercase snake_case throughout. The naming is predictable and makes the action type immediately clear.
14 tools is well within the ideal range for a browser automation server. Each tool covers a meaningful interaction or observation primitive without unnecessary bloat.
The set covers the full core workflow: open, search, observe, interact, extract, screenshot, and close. Minor gaps like explicit back/forward/refresh or tab management exist but are workable through the provided tools.