Fuel Any Agent

Gasoline is an open-source MCP server that implements the Model Context Protocol — a standard for connecting AI assistants to external tools. Any MCP-compatible tool can tap into your browser state. Zero dependencies. Localhost only.

Supported Tools

Tool	Config Location	Guide
Claude Code	.mcp.json (project root)	Setup →
Cursor	~/.cursor/mcp.json	Setup →
Windsurf	~/.codeium/windsurf/mcp_config.json	Setup →
Claude Desktop	OS-specific config file	Setup →
Zed	~/.config/zed/settings.json	Setup →
Gemini CLI	~/.gemini/settings.json	Setup →
OpenCode	~/.config/opencode/opencode.json	Setup →
Antigravity	~/.gemini/antigravity/mcp_config.json	Setup →
VS Code + Continue	~/.continue/config.json	Below

How MCP Mode Works

Gasoline runs as a dual-mode server by default:

• HTTP server — background daemon for the browser extension
• stdio transport — MCP JSON-RPC over stdin/stdout
• Auto-managed — your AI tool starts and stops the server

What MCP tools does Gasoline provide?

Gasoline exposes 5 tools — each with multiple sub-modes controlled by a single parameter.

Tool	What it does	Key sub-modes
observe	Passive browser state	errors, logs, network_waterfall, network_bodies, websocket_events, websocket_status, actions, vitals, page, tabs, screenshot, timeline, error_bundles, saved_videos, recordings, recording_actions, playback_results, log_diff_report
analyze	Active analysis	dom, performance, accessibility, security_audit, third_party_audit, error_clusters, history, link_health, link_validation, page_summary, api_validation, annotations, annotation_detail, draw_history, draw_session
generate	Code and report generation	reproduction, test, pr_summary, sarif, har, csp, sri, visual_test, annotation_report, annotation_issues, test_from_context, test_heal, test_classify
configure	Session management	store, load, noise_rule, clear, health, streaming, recording_start, recording_stop, playback, log_diff, telemetry, diff_sessions, audit_log, describe_capabilities
interact	Browser control	navigate, click, type, select, check, key_press, execute_js, highlight, refresh, back, forward, new_tab, upload, draw_mode_start, screen_recording_start, screen_recording_stop, paste, save_state, load_state, get_readable, get_markdown, navigate_and_wait_for, fill_form_and_submit, run_a11y_and_export_sarif

observe

Read passive browser state. Use the what parameter to select:

Mode	Returns
errors	Console errors with deduplication and noise filtering
error_bundles	Pre-assembled debugging context per error (error + network + actions + logs)
logs	All console output (configurable level and limit)
network_waterfall	Resource timing from PerformanceObserver
network_bodies	Request/response payloads for API debugging
websocket_events	WebSocket messages — filter by connection ID, direction
websocket_status	Active WebSocket connections with message rates
actions	User interactions (click, input, navigate, scroll, select)
vitals	Core Web Vitals — FCP, LCP, CLS, INP
page	Current page URL and title
tabs	All browser tabs with URLs and titles
screenshot	Viewport screenshot
timeline	Merged session timeline (actions + network + errors)
saved_videos	Recorded browser session videos
recordings	Recording metadata
recording_actions	Actions captured during a recording
playback_results	Results from a recording playback
log_diff_report	Compare error states between recordings

analyze

Trigger active analysis. Use the what parameter to select:

Mode	Returns
dom	Live DOM queries with CSS selectors
performance	Performance snapshots with regression detection
accessibility	WCAG audit (axe-core)
error_clusters	Deduplicated error grouping
history	Navigation history
security_audit	Security checks (credentials, PII, headers, cookies)
third_party_audit	Third-party script and origin analysis
link_health	Browser-based link checker with CORS detection
link_validation	Server-side URL validation
page_summary	Page structure summary
api_validation	API contract validation from captured traffic
annotations	Draw mode annotations from user feedback
annotation_detail	Full computed styles for an annotation
draw_history	List past draw mode sessions
draw_session	Get a specific draw session

generate

Generate artifacts from your session. Use the format parameter:

Format	Output
reproduction	Playwright script reproducing user actions
test	Playwright test with network/error assertions
pr_summary	Markdown performance impact summary
sarif	SARIF accessibility report (standard format)
har	HTTP Archive export
csp	Content Security Policy from observed origins
sri	Subresource Integrity hashes for scripts/stylesheets
visual_test	Visual regression test from annotation session
annotation_report	Report from draw mode annotations
annotation_issues	Extracted issues from annotations
test_from_context	Test generated from error/interaction/regression context
test_heal	Repair broken Playwright selectors
test_classify	Classify test failures

configure

Manage session state and settings. Use the action parameter:

Action	Effect
store / load	Persistent key-value storage (save/load/list/delete/stats)
noise_rule	Manage noise filtering rules (add/remove/list/auto_detect)
clear	Clear buffers (network, websocket, actions, logs, all)
health	Server health and memory stats
streaming	Enable/disable real-time event streaming
recording_start	Start capturing a browser session
recording_stop	Stop recording
playback	Replay a recording
log_diff	Compare error states between recordings
telemetry	Configure telemetry metadata mode
diff_sessions	Capture and compare session snapshots
audit_log	View MCP tool usage history
describe_capabilities	List available actions and capabilities

interact

Control the browser. Use the action parameter:

Action	Effect
navigate	Navigate to a URL
click	Click an element (CSS or semantic selector)
type	Type text into an element
select	Select a dropdown option
check	Check/uncheck a checkbox
key_press	Press a key (Enter, Tab, Escape, etc.)
execute_js	Run JavaScript in the page context
highlight	Highlight a DOM element
refresh	Refresh the current page
back / forward	Browser history navigation
new_tab	Open a new tab
upload	File upload for native file dialogs
draw_mode_start	Activate visual annotation overlay
screen_recording_start / screen_recording_stop	Start/stop video recording
paste	Paste text at current focus
save_state / load_state	Save and restore browser state snapshots
get_readable / get_markdown	Extract page content as readable text or markdown
navigate_and_wait_for	Navigate to URL and wait for a selector to appear
fill_form_and_submit	Fill multiple form fields and submit in one action
run_a11y_and_export_sarif	Run accessibility audit and export SARIF in one step

Custom Port

If port 7890 is occupied:

{
  "mcpServers": {
    "gasoline": {
      "command": "npx",
      "args": ["gasoline-mcp", "--port", "7891"]
    }
  }
}

Update the extension’s Server URL in Options to match.

VS Code with Continue

Add to ~/.continue/config.json:

{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "transport": {
          "type": "stdio",
          "command": "npx",
          "args": ["-y", "gasoline-mcp"]
        }
      }
    ]
  }
}

How do I verify Gasoline is connected?

1. Restart your AI tool
2. Gasoline server ignites automatically
3. Extension popup shows “Connected”
4. Ask your AI: “What browser errors do you see?”