dexto
A coding agent and general agent harness for building and orchestrating agentic applications.
claude mcp add --transport stdio truffle-ai-dexto npx -y @anthropics/mcp-server-puppeteer
How to use
Dexto exposes an MCP (Model Context Protocol) server that lets you connect tools, files, and APIs through a standard stdio-based MCP interface. This configuration-driven approach lets you wire up multiple MCP servers (for example, a filesystem server and a browser automation server) and have your agents or external clients talk to them via the MCP protocol without embedding tool logic directly into your code. Within Dexto, you can manage sessions, memory, and tool approvals, and you can browse or add MCP servers from the Web UI or via the CLI. When used as an MCP server, Dexto serves as the orchestration layer that coordinates tools, memory, and agent actions through the MCP stack, enabling multi-tool workflows and cross-server coordination.
How to install
Prerequisites:
- Node.js v14+ (recommend Node.js 18+ for best compatibility)
- npm (or pnpm/yarn)
Install:
- Install Dexto globally
npm install -g dexto
- Start the MCP server mode (Dexto will expose MCP servers over stdio when configured)
dexto --mode mcp
- If you also want a Web UI or CLI for management, install additional tooling as needed (the project provides the web UI/CLI alongside the MCP server configuration).
Notes:
- Ensure your environment allows the MCP configuration to spawn the internal MCP servers (filesystem, browser, etc.) as defined in your mcp_config.
- Logs are useful for troubleshooting; you can configure verbose logging via environment variables as documented by Dexto.
Additional notes
Tips and notes:
- The MCP server in Dexto runs over stdio; your clients connect through the Model Context Protocol using the configured mcpServers entries.
- You can define multiple MCP servers (e.g., filesystem, puppeteer/browser) and reference them by name from your agents' YAML configs.
- Logs are typically found in your home directory under dexto/logs; set DEXT0_LOG_LEVEL=debug for verbose output during troubleshooting.
- If you modify mcpServers, restart the MCP server so changes take effect.
- The MCP integration works with the Web UI and CLI, so you can browse available MCP servers, add new ones from the MCP Store, and monitor connections in real time.
Related MCP Servers
klavis
Klavis AI (YC X25): MCP integration platforms that let AI agents use tools reliably at any scale
aci
ACI.dev is the open source tool-calling platform that hooks up 600+ tools into any agentic IDE or custom AI agent through direct function calling or a unified MCP server. The birthplace of VibeOps.
llm-functions
Easily create LLM tools and agents using plain Bash/JavaScript/Python functions.
web-agent-protocol
🌐Web Agent Protocol (WAP) - Record and replay user interactions in the browser with MCP support
MakeMoneyWithAI
A list of open-source AI projects you can use to generate income easily.
mcp-typescript-sdk
A TypeScript SDK for implementing Model Context Protocol (MCP) over MQTT, supporting both browser and Node.js environments.