Today we’re excited to announce the first release of Nanobot—an open-source framework for turning MCP servers into full-featured AI agents.
Nanobot makes it simple to wrap any MCP server with reasoning, system prompts, tool orchestration, and rich MCP-UI support. Wth Nanobot, developers can build agents that don’t just expose functions, but deliver interactive, engaging user experiences.
Why we built it
If you’ve worked with the Model Context Protocol (MCP), you know how powerful it is. It provides a standard way to connect language models to external tools and data sources. But most MCP servers today are limited to exposing functions: “call this tool, get this data.” Useful, yes. Transformative? Not quite.
We asked ourselves: what if MCP servers could be more than just functions? What if they could be full agents—with personalities, reasoning, and user interfaces—ready to be dropped into any chat client or workflow?
That’s the problem Nanobot was built to solve.
From tools to agents
To make this concrete, imagine you’ve built a Blackjack MCP server. In its simplest form, it might expose tools like “deal,” “hit,” and “stay.” That’s enough for a model to technically play Blackjack, but it doesn’t feel like sitting at a table with a dealer.
Now wrap that server with Nanobot. Suddenly it can:
- Carry a system prompt so it acts like a dealer who explains the rules and offers strategy tips.
- Maintain context across multiple rounds of play.
- Render a full Blackjack table directly inside chat using MCP-UI.
- Respond across channels like Slack or SMS, not just in one interface.
What was once a list of API calls becomes a real interactive experience. To demonstrate, we built HitBot-3000, a blackjack agent running on Nanobot.
Why this matters
The Blackjack example is fun, but the same idea applies across domains.
For an e-commerce company, a Nanobot agent could know the product catalog, place orders, process returns, and also show carts, product images, or order histories directly in the chat window.
For support, an agent could integrate with internal systems, guide a user through troubleshooting, and render dashboards or forms inline.
For internal tools, a Nanobot agent could be a conversational frontend to APIs, while still providing buttons, graphs, or interactive elements when they’re needed.
Instead of bouncing between chat and dashboards, or conversation and control panels, Nanobot lets those experiences converge in one flow. The agent talks to you, acts on your behalf, and shows you what’s happening—all in the same place.
How to use it
Getting started with Nanobot is straightforward:
- Start with an MCP server that exposes your tools.
- Define a Nanobot agent configuration with a system prompt and tool orchestration.
- Add UI components with MCP-UI for an interactive experience.
- Run your agent in the Nanobot host or connect it to any MCP client.
Here’s a simplified example config for wrapping that Blackjack server (you can view a more detailed version on github):
// nanobot.yaml
agents:
dealer:
name: HitBot-3000
model: gpt-4.1
instructions: |
You are a Robot, that is an experienced blackjack dealer who has worked in the finest casinos in the world.
mcpServers: blackjackmcp
mcpServers:
blackjackmcp:
url: https://blackjackmcp.nanobot.ai/mcp
With just a bit of configuration, you’ve transformed a simple MCP server into a full conversational agent that can reason, act, and render UI.
Getting involved
Nanobot is still early, but we believe it’s an important step in the evolution of MCP. MCP gave us a way to connect models to tools. Nanobot makes it possible to turn those tools into intelligent, interactive agents.
We’d love for you to try it out, share feedback, and contribute. You can read more at Nanobot.ai, explore the code and docs on GitHub, read through the examples, or join the community on Discord to see what others are building.
This is just the beginning, and we’re excited to explore what’s possible together.
