Quickstart

Three steps to a working agent. Pick your language once — the tabs stay in sync across the page.

1. Install the package.

   JavaScript

   npm install toolnexus

   Python

   pip install toolnexus

   Go

   go get github.com/muthuishere/toolnexus/golang

   Java

   <!-- Maven -->
   <dependency>
     <groupId>io.github.muthuishere</groupId>
     <artifactId>toolnexus</artifactId>
     <version>0.6.0</version>
   </dependency>

   C#

   dotnet add package Toolnexus

2. Create a toolkit and a client. The toolkit starts with 10 built-in tools (bash / read / write / edit / grep / glob / …). Point the client at any OpenAI- or Anthropic-style endpoint.

   JavaScript

   import { createToolkit, createClient } from "toolnexus"
   
   const tk = await createToolkit()
   const agent = createClient({
     baseUrl: "https://openrouter.ai/api/v1",
     style: "openai",
     model: "openai/gpt-4o-mini",
   })

   Python

   from toolnexus import create_toolkit, create_client
   
   tk = await create_toolkit()
   agent = create_client(
       base_url="https://openrouter.ai/api/v1", style="openai",
       model="deepseek/deepseek-chat",
   )

   Go

   tk, _ := toolnexus.CreateToolkit(ctx, toolnexus.Options{})
   defer tk.Close()
   agent := toolnexus.CreateClient(toolnexus.ClientOptions{
       BaseURL: "https://openrouter.ai/api/v1",
       Style:   "openai",
       Model:   "openai/gpt-4o-mini",
   })

   Java

   Toolkit tk = Toolkit.create(new Toolkit.Options());
   LlmClient agent = LlmClient.create(new LlmClient.Options()
           .baseUrl("https://openrouter.ai/api/v1").style("openai").model("openai/gpt-4o-mini"));

   C#

   await using var tk = await Toolkit.CreateAsync(new Toolkit.Options());
   var agent = LlmClient.Create(new LlmClient.Options
   {
       BaseUrl = "https://openrouter.ai/api/v1",
       Style   = "openai",
       Model   = "anthropic/claude-3.5-sonnet",
   });

3. Run. Tools are called for you and the loop runs to a final answer.

   JavaScript

   const { text } = await agent.run("What files are in this folder?", { toolkit: tk })
   console.log(text)

   Python

   res = await agent.run("List the files here, then count them.", tk)
   print(res.text)

   Go

   res, _ := agent.Run(ctx, "List the Go files here and count them.", tk)
   fmt.Println(res.Text)

   Java

   System.out.println(agent.run("List the files here, then read the largest one.", tk).text);

   C#

   var res = await agent.RunAsync("List the files here, then summarise the README.", tk);
   Console.WriteLine(res.Text);

Add MCP servers and agent skills

Give the toolkit an mcp.json and a skills/ folder and every server tool + a skill tool appear in the same registry:

JavaScript

const tk = await createToolkit({
  mcpConfig: "./mcp.json",     // local stdio + remote streamable-HTTP servers
  skillsDir: "./skills",       // every SKILL.md becomes loadable via one `skill` tool
})

Python

tk = await create_toolkit(mcp_config="./mcp.json", skills_dir="./skills")

Go

tk, _ := toolnexus.CreateToolkit(ctx, toolnexus.Options{
    MCPConfig: "./mcp.json",
    SkillsDir: "./skills",
})

Java

Toolkit tk = Toolkit.create(new Toolkit.Options()
        .mcpConfig("./mcp.json").skillsDir("./skills"));

C#

await using var tk = await Toolkit.CreateAsync(new Toolkit.Options
{
    McpConfig = "./mcp.json",
    SkillsDir = "./skills",
});

Go deeper on each source: MCP servers · Agent skills.

Tip

Next: see all five tool sources — a function, MCP, a skill, a remote A2A agent, and a human-in-the-loop question — in one demo.