Give your agent a body for the web
VidMee is the Visual Agent Runtime: one MCP endpoint (and a plain REST API) that adds 49 tools for stateful browsing, perception, visual QA, memory, and generation to any agent, in any harness, on any LLM. This is the reference. Start with the hero tools, drop to specialists and raw browser driving when you need to.
MCP setup →
Wire VidMee into Claude Code, Cursor, Cline, KiloCode, opencode, Hermes or OpenClaw. One config block, or the one-line installer.
REST API →
No MCP? Every tool is a plain POST /v1/jobs. Auth, the job envelope, polling, idempotency, and errors.
Tool reference →
All 49 tools grouped by layer and namespace, with parameters and examples. The full surface.
Personas →
Four agent lanes (UI/UX, QA, marketing, web assistant): the exact tool stack each one uses, plus recommended add-ons.
The three layers
Reach for the highest layer that answers your question.
| Layer | Use it for | Tools |
|---|---|---|
| Hero | Start here. Each fuses many primitives into one verdict, so one call replaces five. | edit_image inspect report storyboard_video sweep_site test_flow verify_against_reference |
| Specialist | One focused job (render, compare, search, generate, ...). | most of the surface |
| Low-level | Raw stateful browser driving when you script a bespoke flow yourself. | browser_* |
Quickstart
// 1. add the MCP server (Claude Code / Cursor / any HTTP-MCP harness) { "mcpServers": { "vidmee": { "type": "http", "url": "https://mcp.vidmee.ai/mcp", "headers": { "Authorization": "Bearer ${VIDMEE_API_KEY}" } } } } // 2. one call, one verdict inspect({ source: "https://staging.acme.app/checkout", goal: "the Pay button is visible and the form has no errors" })