ChatKnot 🚀

A privacy-focused React Native mobile application for chatting with AI providers and connecting MCP/OpenAPI tools using the Model Context Protocol.

Project Status

• Initial chat app
• MCP support
• Multiple mode support
  • each mode has its own system prompt + model + mcp configuration
• Export chat (markdown/pdf/json)
• AI Search and AI Tab rename
• iamge/file support
• add custom resoning effort/other parameter in the app

🌟 Key Features

• Multi-Provider AI: Configure any OpenAI-compatible endpoint with streaming support.
• MCP Tool Calling: Connect MCP servers and OpenAPI endpoints to extend AI capabilities.
• Multimodal Input: Attach images (for vision-capable models) and text-based documents (PDF, code files, markdown).
• Model Capabilities: Visual capability tags (vision, tools, file) shown next to each model.
• User-Friendly Errors: API errors are parsed and displayed as clean, readable messages.
• Encrypted Storage: Sensitive data and keys are protected with hardware-backed encryption.
• Health Checks: Automatic verification of AI and MCP endpoints on startup.
• Chat Export: Export conversations as PDF (rendered markdown), Markdown (collapsible tool details), or JSON (OpenAI format).
• Clean UI: Modern dark/light interface with organized conversation management.

🛠️ Getting Started

Installation

npm install

Run (Development)

npx expo start

⚙️ Configuration

AI Providers

In Settings -> AI Providers:

1. Add your provider's Base URL (OpenAI-compatible) and API Key.
2. Fetch models and use the Eye icon to make them visible in the chat interface.
3. Edits are handled in Draft Mode — remember to save your changes!

MCP Servers

In Settings -> MCP Servers:

1. Add your MCP server URL (supports OpenAPI discovery).
2. Configure Auto-Approve or Manual mode for tool executions.
3. The app will validate your endpoint spec before saving.

📝 System Prompts

Chat requests use two system instructions:

1. User instruction (global or conversation override prompt)
2. Application default instruction (app behavior, markdown response preference, tool guidance)

MCP/OpenAPI instruction context is only appended in the app instruction when at least one MCP server is connected.

📦 Build Release

To generate a production APK for Android:

npm run android:apk:release

---

ChatKnot is an open-source tool for personal AI productivity.