https://github.com/chuanqisun/react-agent-hooks
Turn React hooks into LLM tools
https://github.com/chuanqisun/react-agent-hooks
genui library llmstxt mcp react react-hooks
Last synced: 5 months ago
JSON representation
Turn React hooks into LLM tools
- Host: GitHub
- URL: https://github.com/chuanqisun/react-agent-hooks
- Owner: chuanqisun
- Created: 2025-04-09T14:54:43.000Z (6 months ago)
- Default Branch: master
- Last Pushed: 2025-04-28T20:32:25.000Z (6 months ago)
- Last Synced: 2025-04-28T20:45:54.283Z (6 months ago)
- Topics: genui, library, llmstxt, mcp, react, react-hooks
- Language: TypeScript
- Homepage:
- Size: 994 KB
- Stars: 71
- Watchers: 2
- Forks: 2
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
Awesome Lists containing this project
- awesome-ccamel - chuanqisun/react-agent-hooks - Turn React hooks into LLM tools (TypeScript)
README
# React Agent Hooks
| Agentic Counter Demo | Agentic Todo Demo |
| :---------------------------------------------------------------------------------------------------------------------------------------------------------- | :------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| [](https://stackblitz.com/edit/react-agentic-counter?file=src%2Fmain.jsx) | [](https://stackblitz.com/github/chuanqisun/react-agent-hooks?file=src%2Fmain.jsx) |
Turn React Hooks into LLM Tools
- 🪝 Familiar: same semantics as React hooks
- 🤝 Symbiotic: human interface and Agent interface derived from the same state.
- 🛡️ Safe: developer controls the schema for Agentic state change.
- ➕ Incremental adoption: use as much or as little as you want.
- 📦 Composable: fully interoperable with classic React hooks.
- 🔮 Future-ready: forward-compatible with MCP and llms.txt.
**Before**
```jsx
import { useCallback, useState } from "react";
function MyComponent() {
const [name, setName] = useState("John Doe");
const [age, setAge] = useState(30);
const adjust = useCallback((delta) => setAge((prev) => prev + delta), []);
return (
{name}
{age}
adjust(-1)}>Be younger
adjust(1)}>Be older
);
}
```
**After**
```jsx
import { useAgent, useAgentState, useAgentTool } from "react-agent-hooks";
export function MyComponent() {
const agent = useAgent({ apiKey: "******" });
const [name, setName] = useAgentState("Name", "John Doe");
const [age, setAge] = useAgentState("Age", 30);
const adjust = useCallback((delta) => setAge((prev) => prev + delta), []);
useAgentTool("Change age", z.number().describe("the delta of age change"), adjust);
return (
{name}
{age}
agent.run("be younger")}>Be younger
agent.run("be older")}>Be older
);
}
```
[](https://stackblitz.com/edit/react-agentic-counter?file=src%2Fmain.jsx)
## Get Started
```sh
npm install react-agent-hooks
```
## Usage
### Give Agent "Eyes" 👀
```tsx
import { useAgentMemo } from "react-agent-hooks";
function MyComponent() {
const [name, setName] = useState("John Doe");
const [age, setAge] = useState(30);
// Describe a readable state to the Agent
useAgentMemo("User's profile", () => ({ name, age }), [name, age]);
return (
{name}
{age}
);
}
```
### Give Agent "Hands" 👏
```tsx
import {z} from "zod";
import { useAgentState, useAgentTool } from "react-agent-hooks";
function MyComponent() {
// Describe a readable state to the Agent while exposing a setter function to developer
const [foodPreferences, setFoodPreferences] = useAgentState("food preference", ["Pizza", "Sushi"]);
// Wrap the setter as a tool and describe it to the Agent
const addFoodPreference = useAgentTool("add-food-preference", z.object(foodItems: z.array(z.string())), (foodItems) => {
setFoodPreferences((prev) => [...prev, ...foodItems]);
});
const removeFoodPreference = useAgentTool("remove-food-preference", z.object(foodItems: z.array(z.string())), (foodItems) => {
setFoodPreferences((prev) => prev.filter((item) => !foodItems.includes(item)));
});
return
- {item} )}
{foodPreferences.map(item =>
}
```
### Run the Agent
```tsx
import { useAgent } from "react-agent-hooks";
function MyApp() {
// Run the Agent with a prompt
// Agent always sees the latest states from `useAgentState`, `useAgentMemo`, and can uses the latest tools from `useAgentTool`
const agent = useAgent({ apiKey: "******" });
// Call the Agent
const handleFormSubmit = (e) => {
e.preventDefault();
const input = e.target.elements[0].value;
agent.run(input);
};
return (
Ask Agent
);
}
```
### Compose Agentic Application
Inside a component, use the `enabled` option to dynamically show/hide states and tools to the Agent.
```tsx
const shouldShowFeature = true; // You can dynamically decide this value
useAgentMemo("User's profile", () => ({ name, age }), [name, age], { enabled: shouldShowFeature });
useAgentState("some state", { name: "Some state" }, { enabled: shouldShowFeature });
useAgentTool(
"update state",
z.object({ name: z.string() }),
(newState) => {
setSomeState(newState);
},
{ enabled: shouldShowFeature },
);
```
In a component tree, use JSX to dynamically show/hide states and tools to the Agent.
```tsx
function ParentComponent() {
// A higher level component can dynamically decide what lower level states/tools are available
const = [shouldShowFeature, setShouldShowFeature] = useAgentState("toggle feature", z.boolean(), true);
useAgentTool("toggle feature", z.object({}), () => setShouldShowFeature(prev) => !prev);
return {shouldShowFeatureB ? : null};
}
function ChildComponent() {
// The state and tool will be available to the Agent only if the child component is rendered
useAgentState("some state", { name: "Some state" });
useAgentTool("update state", z.object({ name: z.string() }), (newState) => {
setSomeState(newState);
});
return
}
```
### Build a custom Agent
Access currently active states and tools with `useAgentContext` hook. Here is an example of building your own agent
```tsx
export function useMyAgent() {
const openai = new OpenAI({ dangerouslyAllowBrowser: true, apiKey: "******" });
const agentContext = useAgentContext();
const run = async (prompt: string) => {
const task = openai.beta.chat.completions.runTools({
stream: true,
model: "gpt-4.1",
messages: [
{
role: "system",
content: `
User is interacting with a web app in the following state:
\`\`\`yaml
${agentContext.stringifyStates()}
\`\`\`
Based on user's instruction or goals, either answer user's question based on app state, or use one of the provided tools to update the state.
Short verbal answer/confirmation in the end.
`.trim(),
},
{
role: "user",
content: prompt,
},
],
tools: agentContext.getTools(),
});
return task;
};
return {
run,
};
}
```
### Scale-up with Context
The `AgentContext` is an optional React Context to help you hierarchically organizing states and tools.
This prevents naming collisions and reduces agent confusion from too many similar states and tools.
```tsx
import { AgentContext } from "react-agent-hooks";
function MyApp() {
return (
);
}
function HeaderContent() {
// The Agent will see this state appear within the "app root > nav > header" context
const [someState, setSomeState] = useAgentState("some state", { name: "Some state" });
return
}
```
## Future Work
Render to MCP Server
```tsx
import { renderToMCPServer } from "react-agent-hooks";
function main() {
// Spin up an MCP server at port 3000.
// React Agent state -> MCP resource and prompts
// React Agent tools -> MCP tools
const server = renderToMCPServer().listen(3000);
}
```
Render to llms.txt
```tsx
import { renderToLlmsTxt } from "react-agent-hooks";
function main() {
server.get("/llms.txt", (req, res) => {
const userContext = req.query.userContext;
const llmsTxtContent = renderToLlmsTxt();
res.send(llmsTxtContent);
});
}
```
## Reference
Blog article: [React (hooks) is All You Need](https://stackdiver.com/posts/react-hooks-is-all-you-need/)