Jev Browser
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| JEV_PROVIDER | No | Force Jev provider: 'typesafe' or 'openrouter' instead of auto-detection. | auto |
| GEMINI_API_KEY | No | Google Gemini API key (starting with AIza) for the typing model. | |
| OPENAI_API_KEY | No | OpenAI API key (starting with sk-) for the typing model. | |
| TYPESAFE_API_KEY | No | TypeSafe API key. Default provider when set. Obtainable from console.typesafe.ai/settings/keys. | |
| ANTHROPIC_API_KEY | No | Anthropic API key (starting with sk-ant-) for the typing model. | |
| JEV_BROWSER_MODEL | No | Pin a Jev version, or 'typesafe/jev-1.13' on OpenRouter. | jev-latest |
| JEV_BROWSER_HEADED | No | Set to '1' to watch the browser. | |
| OPENROUTER_API_KEY | No | OpenRouter API key. Powers both Jev judgments and the typing model if no other typing provider is configured. One key runs everything. | |
| JEV_BROWSER_TYPE_MODEL | No | Override the typing model, e.g. 'openrouter:anthropic/claude-haiku-4.5'. | |
| JEV_BROWSER_TYPE_API_KEY | No | API key for the typing endpoint if required. | |
| JEV_BROWSER_TYPE_BASE_URL | No | Base URL for any OpenAI-compatible endpoint for the typing model (e.g. Ollama, LM Studio). | |
| JEV_BROWSER_TYPE_PROVIDER | No | Force typing provider: 'openai', 'openrouter', 'anthropic', or 'google'. | |
| GOOGLE_GENERATIVE_AI_API_KEY | No | Alternate Google API key for the typing model. | |
| JEV_BROWSER_SKIP_BROWSER_DOWNLOAD | No | Set to '1' to skip the Chromium postinstall download. |
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 |
|---|---|
| jev_navigateA | Give a task and a start URL; a Jev-driven agent navigates a real headless browser until the goal is met, the stuck gate fires, or a budget (steps/seconds) is exhausted. Returns the final page in a chosen format (text, markdown, html, or an aria snapshot), the full step trace with confidences, console/page/network errors captured along the way, token usage with estimated cost, and a final screenshot. |
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 1 tool
With only a single tool, there is no possibility of confusion between different tools. The tool's purpose is clear and distinct, and an agent can easily select it for any browser navigation task.
There is only one tool name, 'jev_navigate', which follows a clean verb_noun pattern. Since there are no other tools to compare against, naming is inherently consistent.
A single tool is on the thin side for a browser server, but the tool is comprehensive, handling navigation, tracing, error capture, and screenshot generation. It is borderline but acceptable for a focused high-level automation agent.
The tool covers the core navigation workflow and returns extensive diagnostic data, but it lacks granular browser operations like clicking, typing, or extracting specific elements. Agents requiring fine-grained control would need to rely entirely on this one high-level action, which is a notable gap.