Support

Install from the Chrome Web Store. One click and the FSB icon appears in your toolbar. Prefer to run from source? Download the latest release from GitHub, then load it unpacked: open chrome://extensions, enable Developer Mode, click "Load unpacked," and select the FSB folder.

Choose a supported provider and model in Options based on your task, budget, and privacy needs. Use your own API key for hosted models, OpenRouter for access to multiple providers, or LM Studio for local models. Check your provider's current availability and pricing before choosing a model.

FSB requires Chrome 116 or later, or a compatible desktop Chromium browser such as Microsoft Edge, Brave, Opera, or Vivaldi. Firefox is not currently supported.

API keys are encrypted with AES-GCM in Chrome's local storage. Requests to hosted AI models send task data to your selected provider. Vault tools can fill credentials and payment details in the browser without sending raw secrets across the MCP bridge.

FSB includes action validation and safety checks. It can work across multiple tabs; MCP agents use agent identity and tab ownership to scope actions. Autopilot runs when you explicitly delegate a task. Review automation involving sensitive accounts or payments. The source code is open and auditable.

Open an issue on the GitHub Issues page at https://github.com/fullselfbrowsing/FSB/issues. Include the website URL, your task description, and any error messages from the FSB chat or browser console. Session logs from the Debug tab in Options can also help.

MCP (Model Context Protocol) lets you control FSB from AI coding tools like Claude Code, Codex, Cursor, Windsurf, and OpenClaw. The MCP server runs as npx -y fsb-mcp-server@latest and pairs with the running extension over a local WebSocket bridge. FSB exposes 68 MCP tools (click, type, navigate, read DOM, fill vault credentials, and more). For OpenClaw users specifically, the fastest path is the one-click install on ClawHub at https://clawhub.ai/lakshmanturlapati/full-selfbrowsing, or load the FSB skill at skills/fsb/ which prints the stdio config block and runs the doctor diagnostic. The full rundown lives at https://full-selfbrowsing.com/agents.

Connect FSB as an MCP server, then let the coding agent open the local app, click through real flows, observe DOM state, capture console or task errors, and report failures back into the coding session. This closes the manual loop between code generation, browser testing, logging, and fixing.

Yes. The Remote Dashboard at full-selfbrowsing.com lets you pair with your FSB extension via QR code, then send tasks, view a live DOM preview of the browser, and monitor current execution state from any device with a web browser.

FSB has a local memory system with three layers: semantic memory (site navigation patterns), episodic memory (past action sequences), and procedural memory (successful workflows). This data stays on your device and helps FSB navigate familiar sites faster and avoid repeating mistakes. Memory auto-consolidates after enough sessions.

Run npx -y fsb-mcp-server@latest install --list to see which MCP hosts the installer detects on your machine, then run the installer for one (for example npx -y fsb-mcp-server@latest install --claude-desktop). Cursor, Codex, Windsurf, and others have matching flags from --list. For OpenClaw, paste the canonical stdio block printed by the FSB skill (loaded from skills/fsb/) into your OpenClaw MCP config -- the bare --openclaw installer flag is intentionally manual because OpenClaw's MCP config schema is still unstable across builds. Verify the install with npx -y fsb-mcp-server@latest doctor; six green layers means you're ready.

FSB exposes two modes through MCP. Manual mode is the default and gives the calling agent 68 granular tools: navigate, click, type, scroll, read page state, take screenshots, upload real local files with upload_file, arm trigger watchers with trigger / get_trigger_status / list_triggers, and search or invoke guarded first-party API capabilities with search_capabilities and invoke_capability. Vault and payment helpers fill inside the encrypted extension store, so raw secrets do not cross the MCP bridge. Autopilot mode (run_task) only fires when the user explicitly delegates the whole task. For everything else the calling agent stays in the loop on every action.

Three options, fastest first. One-click on ClawHub: install from https://clawhub.ai/lakshmanturlapati/full-selfbrowsing. FSB skill: load the skill from skills/fsb/ in the FSB repo. It prints the canonical OpenClaw stdio block for you to paste into OpenClaw's MCP config and offers consent-gated install for any other MCP hosts on your machine. The full 3-step walkthrough (Chrome extension, MCP server config, doctor verification) lives at https://full-selfbrowsing.com/agents.

Run npx -y fsb-mcp-server@latest doctor. The doctor walks six layers -- package, bridge, extension, active-tab, content-script, config -- and exits non-zero on the first failure with a one-line next-step recommendation. The full recovery table for each layer lives in skills/fsb/USAGE.md, and the same diagnostic runs automatically when the FSB skill is loaded into OpenClaw.

FSB is free and open source under the MIT license. When you use your own API key, your chosen provider sets and bills model usage. Agent CLI mode uses your existing agent setup; MCP mode uses the calling agent's model. Local models through LM Studio do not require a hosted-provider API key.