Superpower

Bring MCP tools into the AI web apps you already use.

Run local and remote MCP tools from ChatGPT, Gemini, Perplexity, Grok, GitHub Copilot and other supported web assistants without leaving the conversation.

Quick start · Features · Supported platforms · Architecture · Development

What is Superpower?

Superpower is a Chrome extension that connects supported AI web interfaces to the Model Context Protocol through an MCP proxy. It detects structured tool calls in the conversation, routes them to connected MCP servers, executes the requested tools, and renders the results back into the same chat workflow.

The browser UI stays where you already work. MCP execution stays behind the local or remote endpoint you configure.

Features

Illustrated sidebar overview based on the current controls.

A tool call from model output to MCP execution and result injection.

Built for real MCP workflows

Superpower supports SSE, WebSocket, and Streamable HTTP connections. Tool instructions are generated from the tools exposed by the connected server, while V1 adds dependency-aware batching guidance and separate function-call blocks to make multi-tool workflows easier for models to follow.

Supported platforms

The current adapter set includes:

Web UI changes can affect DOM-based adapters. If one platform changes its composer or response markup, open an issue with the affected platform and browser version.

How it works

At a high level:

1. A supported AI website produces a structured tool call.
2. Superpower detects and parses the call in the browser.
3. The call is forwarded through the configured MCP connection.
4. The MCP server executes the tool and returns its result.
5. Superpower renders or inserts the result back into the conversation so the model can continue.

Quick start

Requirements

• Node.js 22.12+
• pnpm 9.x
• Chrome or another Chromium-based browser
• One or more MCP servers you want to expose through the proxy

1. Clone and install

git clone https://github.com/stloendays/Superpower-V1.git
cd Superpower-V1
pnpm install

2. Create an MCP proxy configuration

Create config.json outside the repository or in a local ignored path:

{
  "mcpServers": {
    "example-server": {
      "command": "npx",
      "args": ["-y", "your-mcp-server-package"]
    }
  }
}

Do not commit credentials, API keys, access tokens, or private machine paths.

3. Start the MCP proxy

SSE:

npx -y @srbhptl39/mcp-superassistant-proxy@latest \
  --config ./config.json \
  --outputTransport sse

Other supported proxy transports:

# Streamable HTTP
npx -y @srbhptl39/mcp-superassistant-proxy@latest --config ./config.json --outputTransport streamableHttp

# WebSocket
npx -y @srbhptl39/mcp-superassistant-proxy@latest --config ./config.json --outputTransport ws

The proxy package currently uses the mcp-superassistant-proxy package name. Superpower uses it as an external compatibility dependency.

4. Build the extension

pnpm base-build

The unpacked extension is generated in dist/.

5. Load it in Chrome

1. Open chrome://extensions/.
2. Enable Developer mode.
3. Select Load unpacked.
4. Choose the generated dist/ directory.
5. Open one of the supported AI websites and use the Superpower sidebar to connect to your MCP endpoint.

Connection endpoints

The default V1 connection is SSE at http://localhost:3006/sse.

Typical workflow

1. Start the MCP proxy and confirm the desired MCP servers are available.
2. Open a supported AI platform.
3. Connect Superpower to the proxy from the server-status UI.
4. Choose which tools should be exposed to the assistant.
5. Insert or attach the generated MCP instructions when needed.
6. Ask the assistant to perform a task that requires one of the enabled tools.
7. Review the detected function call and run it manually, or use the available automation controls.
8. Continue the conversation with the returned tool result.

V1 highlights

Superpower focuses on a streamlined browser-side MCP workflow:

• Superpower branding and the custom lightning-S extension icon
• Chrome extension version 1.0.0 with display version V1
• Dependency-aware MCP instruction guidance for multi-tool workflows
• Separate function-call blocks in generated instructions
• ChatGPT MCP-file submission refresh handling
• Release-oriented environment templates, security notes, and repository cleanup

See CHANGELOG.md for the release summary.

Development

Development build

pnpm dev

Production build

pnpm base-build

Type checking and linting

pnpm type-check
pnpm lint
pnpm prettier

Firefox build

The monorepo includes a Firefox build path:

pnpm build:firefox

Browser behavior and support should be validated separately before publishing a Firefox release.

Repository structure

Superpower-V1/
├── chrome-extension/          # Manifest V3 extension core and background service
├── pages/content/             # Content UI, adapters, tool rendering, sidebar
├── packages/                  # Shared monorepo packages
├── scripts/                   # Local repository setup helpers
├── docs/readme/               # README visual assets
├── CHANGELOG.md
├── SECURITY.md
└── README.md

Security

MCP servers can expose powerful capabilities such as filesystem access, developer tools, databases, and third-party APIs. Only connect Superpower to endpoints you trust.

In particular:

• Keep local proxy ports private unless you intentionally configure network access and authentication.
• Review the tools exposed by each MCP server before enabling automation.
• Keep secrets in local environment/configuration files that are excluded from Git.
• Treat tool output and file attachments as potentially sensitive data.

See SECURITY.md for project-specific guidance.

Contributing

Issues and pull requests are welcome.

When reporting bugs, please include the affected AI platform, browser version, reproduction steps, and the relevant page behavior. For compatibility-related changes, please keep unrelated refactors separate so that fixes can be reviewed and maintained more effectively.

Author

Superpower is created and maintained by Junbo, from the National University of Singapore (NUS).

The project is independently developed with the goal of building a more powerful, flexible, and seamless interface for connecting AI platforms with external tools and workflows.

License

Released under the MIT License.

Copyright © 2026 Junbo.