MCP Server

Your Agent's Browser.
Native in Every AI Host.

AGENSTAB now speaks the Model Context Protocol. Connect Claude Desktop, Claude Code, Cursor — any MCP-compatible host — to a sovereign browser session with one config block. Fifteen tools. Zero glue code.

Get Your API Key Read the Docs

Claude Desktop·Claude Code·Cursor·Any MCP Host

Two Ways to Integrate

	SDK (Code)	MCP (Config)
Setup	Install SDK, write integration code	Paste one JSON block
Who drives	Your application logic	Your AI assistant, directly
Languages	Python, Node.js, Go	Any MCP-compatible host
First action	Write code	Ask your assistant to browse
Auth & limits	Engine-enforced	Engine-enforced — same key, same tiers

How it works

Get a key

Authenticate and copy your ak_live_ API key from the console

Add the config

One JSON block in your MCP host's settings

Ask it to browse

Your assistant navigates, observes, clicks, and extracts — through the sovereign engine

One Config Block

Claude Desktop Setup

claude_desktop_config.json

{
  "mcpServers": {
    "agenstab": {
      "command": "npx",
      "args": ["-y", "@agens-tab/mcp"],
      "env": {
        "AGENSTAB_API_KEY": "ak_live_your_key_here"
      }
    }
  }
}

Claude Code: claude mcp add agenstab -e AGENSTAB_API_KEY=ak_live_... -- npx -y @agens-tab/mcp

Fifteen Tools

Each maps directly to the engine's browser control surface. Your assistant calls them by name.

navigate

Go to a URL and wait for the page to settle.

observe

Read the page's semantic accessibility tree — the primary perception tool.

click

Click an element by its stable semantic id.

type

Enter text into a field.

select

Choose an option in a dropdown.

scroll

Scroll the page or a scrollable element.

fill_form

Populate multiple fields in one call.

wait_for

Wait for an element to appear.

extract_table

Pull a table off the page as structured rows.

screenshot

Capture the current viewport as an image.

get_dom

Return the page DOM, clean or raw.

get_session_status

Report session health and action count.

get_audit_log

Return the compliance trail for the session.

run_agent

Execute a packaged AGENSTAB agent blueprint by id.

close_session

Terminate the session and free engine resources.

No New Attack Surface

Same key, same limits

The MCP server authenticates with your standard API key and is bound by the same tier limits, RBAC, and quotas the engine already enforces. It has no capability your key doesn't.

Credential isolation

Your key lives only in the host's environment — never in logs, never in the protocol stream. Read-only mode is available for observation-only deployments.

Safe by default

Arbitrary code execution is opt-in twice — a server flag and an engine session setting. The sovereign engine's navigation guard, redaction, and audit logging remain the system of record.

Give your assistant a browser.

Get an API key and connect in under five minutes.

Get Your API Key Read the Docs