Slim Mode

Expose only 5 essential MCP tools to reduce token overhead from ~4,200 to ~600 tokens.

Slim Mode

Start the MCP server with --slim to expose only 5 essential tools instead of 23. This cuts tool definition tokens from approximately 4,200 to approximately 600 — a major context savings for AI agents.

Enable Slim Mode

CLI

npx @browseragentprotocol/mcp --slim

The 5 Essential Tools

Tool	Description
navigate	Navigate to a URL (with fused observe support)
observe	See interactive elements with stable refs
act	Batch multiple interactions in one call
extract	Extract structured data with JSON schema
screenshot	Capture page screenshot

These 5 tools cover 95%+ of browser automation tasks.

What You Lose

With slim mode, these tools are not available:

• Individual interaction tools: click, fill, type, press, select, scroll, hover — use act to batch these instead
• Advanced observation: accessibility, aria_snapshot, content, element — use observe with responseTier
• Page management: pages, activate_page, close_page — single-tab workflows only
• Navigation helpers: go_back, go_forward, reload — navigate to URLs directly
• Discovery: discover_tools — no WebMCP discovery

In slim mode, use act for all interactions. Instead of calling click directly, use:

act({ steps: [{ action: "action/click", selector: "role:button:Submit" }] })

When to Use Slim Mode

Use Slim Mode

• Most browser automation tasks - When token budget is limited - Single-tab workflows - When using act for batched interactions

Use Full Mode

• Multi-tab workflows requiring pages/activate_page - When you need aria_snapshot for cheap page structure - When you need content for markdown extraction - When you need element for state queries - When using WebMCP discover_tools

Token Comparison

Mode	Tools	Definition tokens
Full	23 tools	~4,200 tokens
Slim	5 tools	~600 tokens
Savings		~85%

Start with slim mode. If you hit a limitation, switch to full mode. Most agents never need more than the 5 essential tools.