One MCP layer for browser AI and native desktop workflows.
English · 简体中文
Problem · Products · Tour · Why Superpower · Features · Platforms · Architecture · Quick start · Operation guide · Repository
Practical MCP workflows across browser and desktop surfaces.
Release status: v1.4.1 is the current public stable release. V1.5 is the active development line and adds a browser Copilot workspace, any-page capture, YouTube summarization, local Knowledge capture, connected MCP actions, and review-first workflow execution.
Web AI chats are good reasoning surfaces, but they normally cannot directly use your local files, command-line tools, private MCP servers, or desktop workflows. Superpower connects those two worlds: the web model can decide what tool work is needed, while MCP executes the requested action through a local or remote tool server and returns the result to the same conversation.
That makes Superpower useful when you want to keep working in ChatGPT, Gemini, Perplexity, Grok, Copilot, Qwen or another supported web chat while still using real MCP tools. It is not a replacement for coding agents such as Codex; instead, it provides a browser-first execution layer for users who prefer the web chat as the planning interface, want their own MCP stack, need a review step before execution, or want lightweight tool use without consuming a separate local-agent model quota.
In one line: use the AI chat you already have as the reasoning surface, and use MCP as the execution layer.
|
Bring MCP tools directly into supported AI websites without leaving the conversation.
Install: Chrome Web Store Best for: browser-native AI workflows. |
Guide: Desktop usage and Quick Ask Use Superpower as a native MCP workspace outside the browser.
Best for: managing MCP connections and actions from one desktop workspace. |
Illustrative product concept showing how Superpower can bring research, writing and tool-driven actions into one browser-first workflow. For the current compatibility list, see Supported platforms.
Superpower keeps tool execution close to the interface where the work happens. Use the browser extension when the conversation is the workspace, or use the desktop app when you want a dedicated MCP control surface.
|
Use MCP capabilities without constantly switching tools or consoles. |
Use local or remote MCP servers across browser and desktop workflows. |
Choose exposed tools and review guarded actions before execution. |
- Shared MCP execution model across browser and desktop surfaces
- Browser Copilot workspace for page/selection-aware prompts inside supported AI sites
- Lightweight Page Assistant on ordinary HTTP(S) pages for selection-level Summarize, Explain, Ask AI, Save, page summarization, and action extraction
- Cross-tab AI routing: page actions go to the most recently used supported AI tab, or open ChatGPT when no supported AI workspace is available
- YouTube watch-page summarization and Save to Knowledge support: use loaded transcript segments when available and fall back to supplied video metadata when they are not
- One-click Summarize, Explain, Improve writing, Translate, Action items, and Study notes workflows
- Local Knowledge capture for selected text with page title, source URL, timestamps, source classification, tags, search/filter controls, and Markdown export
- Multi-select Knowledge workflows: attach saved notes to Ask, synthesize multiple sources, export selected items, or hand them to a compatible notes/file MCP tool
- Knowledge Projects: organize the same saved source across multiple research Projects without duplicating content; open a Project to filter its sources, attach the whole Project to Ask, synthesize/export it, or send it through a compatible MCP destination
Ctrl/⌘ + Shift + Kshortcut to surface the Copilot workspace from a supported AI page- Browser-side tool discovery, structured call detection and result injection
- Connected Actions for detected notes, mail, calendar, task, Slack/messaging, and file-storage MCP tools, while preserving the review-first execution path
- Native desktop workspace for connections, apps, actions and runs
- Local and remote MCP endpoints
- Protocol-level MCP health checks, heartbeat-driven recovery, network/foreground recovery, and bounded exponential reconnect backoff
- Connection-aware automation controls for Auto Execute, Auto Insert, and Auto Submit, with configurable delays and tool-catalog refresh after reconnect
- Review-first natural-language Action Router
- Dependency-aware Workflow Planner with explicit cross-step bindings
- Session-only Workflow Runner with one-step-at-a-time guarded execution
- Persistent controls and generated MCP instructions
Illustrative capability overview. Superpower keeps reasoning in the AI surface while MCP provides the execution layer.
The V1.5 development line now starts from a Copilot-first browser workspace. On supported AI sites, Superpower can use the current selection or bounded page context to prepare structured prompts for summarization, explanation, rewriting, translation, action extraction, and study notes, then submit them through the site's existing AI composer. Selected text can also be saved locally into Knowledge with source metadata and exported as Markdown.
Outside the supported AI sites, a lightweight Page Assistant provides a small floating control without loading the full MCP/adapter stack. Selection and page-level tasks are routed to the most recently used supported AI tab; if none is available, Superpower opens ChatGPT and waits for the adapter before submitting. On YouTube watch pages, the assistant can build a video-summary request from the title, channel, description, and transcript segments when the transcript is available in the loaded page.
Knowledge is no longer only a local clip list. Saved items can be searched across title, text, URL and tags, filtered by source type (Web / YouTube / AI chat / Other), tagged, multi-selected and attached directly to the Copilot Ask box. Multiple saved sources can be synthesized with provenance preserved, copied as structured Markdown, or handed to a compatible notes/file MCP tool such as Notion or Drive through the review-first workflow. Knowledge Projects add a non-destructive grouping layer: a source can belong to multiple Projects without duplicating its content, and an entire Project can be attached to Ask, synthesized, exported, or handed to MCP as one research context.
Connected MCP tools are also promoted into a Connected Actions panel. When compatible tools are detected, Superpower can surface actions such as Save note, Draft email, Plan event, Create task, Draft Slack, or Search files. These shortcuts still use the existing review-first MCP path rather than bypassing tool schemas or confirmation rules.
V1.5 also hardens the underlying connection and automation loop. Superpower validates liveness with MCP protocol ping, uses heartbeat failures plus browser network/visibility events to trigger recovery, refreshes the tool catalog after reconnect, and uses bounded exponential backoff instead of waiting only for the periodic checker. Auto Execute performs a connection preflight before clicking Run; Auto Insert returns successful results to the AI composer; Auto Submit only follows a successful insert. Once a tool request may have reached the server, Superpower does not automatically replay it after a timeout or ambiguous failure, avoiding duplicate state-changing actions.
The natural-language Action Router continues to map intent to MCP capabilities, draft schema-backed parameters, plan multi-step workflows, make cross-step data bindings explicit, and keep guarded execution reviewable. Workflow runs advance at most one MCP Action at a time; reviewed bindings and policy checks remain authoritative.
Illustrative workflow gallery: move from reading and synthesis to structured actions while staying in the same working context.
More workflow examples
Understand a page: summarize long-form content, surface key points and keep contextual follow-up questions beside the source.
Research and synthesis: collect sources, compare evidence, extract useful highlights and keep reusable notes in context.
Write and refine: keep the draft visible while MCP-backed actions help rewrite, simplify, improve tone or build an outline.
Capture and act: preserve useful highlights, turn them into tasks, organize them and export the result into the rest of your workflow.
The browser extension currently supports ChatGPT, Google Gemini, Perplexity, Google AI Studio, Grok, OpenRouter, DeepSeek, T3 Chat, GitHub Copilot, Mistral, Kimi, Qwen Chat, and Z.ai.
- A browser or desktop workflow selects an MCP capability.
- Superpower routes the structured request through the configured MCP connection.
- The MCP server executes the tool.
- The result is returned to the active browser conversation or desktop workflow.
- Install Superpower from the Chrome Web Store.
- Open the extension in Chrome and configure your MCP connection.
- For the standard local proxy, use Streamable HTTP with
http://localhost:3006/mcp. - Open a supported AI website. In the V1.5 development line, the sidebar opens to Copilot by default; use a quick action or press
Ctrl/⌘ + Shift + Kto return to it. - On ordinary web pages, use the small Superpower Page Assistant to summarize a selection/page, save content to Knowledge, or route the task into your active AI workspace. YouTube watch pages add a Summarize video action.
- Inside Copilot, create Knowledge Projects to group saved sources by research topic or workstream. Add selected Knowledge to a Project, then use the whole Project in Ask, synthesis, Markdown export, or MCP handoff.
- In Settings → Automation, optionally enable Auto Execute, Auto Insert, and Auto Submit. Auto Submit depends on a successful Auto Insert, and Auto Execute checks MCP connectivity before running.
- Use Connected Actions when Superpower detects compatible MCP tools; switch to Tools when you want the full tool interface.
Connection compatibility: new local setups default to Streamable HTTP. Explicit legacy SSE endpoints such as
http://localhost:3006/sseand WebSocket endpoints remain supported. Existing saved connection settings are not overwritten.
New to Superpower? Follow the Browser Extension Operation Guide for the full connect → inspect tools → review instructions → run a safe first task workflow. The same guide also explains how to test an unpacked local build without publishing it to the Chrome Web Store.
For manual or development installation, see docs/install/windows-extension.md.
- Open the latest release.
- Download
Superpower-Desktop-*-Windows-x64.zip. - Extract the archive and launch the packaged desktop application.
Requirements: Node.js 22.12+, pnpm 9.x, and a Chromium-based browser.
git clone https://github.com/stloendays/Superpower.git
cd Superpower
pnpm install
pnpm base-buildConfigure an MCP proxy, start it with your preferred transport, then load dist/ as an unpacked extension. For the exact reload-and-refresh loop used during development, see Test a local build without publishing.
pnpm dev # Development build
pnpm base-build # Production build
pnpm type-check # Type checking
pnpm lint # LintThe repository root is intentionally kept as an entry surface: standard project documents, workspace/build configuration, and the main product directories stay visible; implementation helpers and optional examples live under dedicated folders.
Superpower/
├── chrome-extension/ # Browser extension shell and background integration
├── desktop/ # Native Qt desktop application and desktop examples
├── docs/ # Documentation, install guides and README/site assets
├── packages/ # Reusable TypeScript packages, including MCP Core/Host
├── pages/ # Browser extension pages and content scripts
├── scripts/
│ ├── install/ # Windows source/release installer helpers
│ └── shell/ # Build, environment and version shell utilities
├── README.md # Project homepage
├── README.zh-CN.md # Simplified Chinese homepage
├── CHANGELOG.md
├── SECURITY.md
└── LICENSE
Superpower executes user-configured MCP tools. Treat connected MCP servers and tool outputs as trusted only to the extent you trust their source and configuration.
- Review the MCP server configuration before connecting.
- Keep credentials out of public configuration files and issue reports.
- Prefer loopback-only local endpoints when remote access is unnecessary.
- Review guarded or state-changing actions before execution.
See SECURITY.md for the current security guidance.
Issues and pull requests are welcome. Please keep changes scoped, document behavior changes, and include validation where practical.
Superpower is relevant to developers and researchers working with Model Context Protocol (MCP), AI agents, browser automation, Chrome extensions, agentic workflows, tool-using LLMs, local MCP servers, desktop AI workflows, Qt 6, and human-in-the-loop execution.
Useful search terms: MCP browser extension, Model Context Protocol Chrome extension, ChatGPT MCP tools, Gemini MCP, browser AI agent, agentic workflow, tool calling, MCP desktop client, MCP Streamable HTTP, human in the loop MCP.
Superpower is released under the MIT License.