AgentHub IDE

Overview

AgentHub IDE is a purpose-built IDE for developing, debugging, and running AI agent systems. It launched September 24, 2026 (196 HN points) as the successor-in-spirit to AgentsDock (covered in the September 15 roundup) but with MCP 2.2-native support and first-class integration with the Agent Skill Protocol v2.

Where AgentsDock targeted researchers building novel agent topologies, AgentHub IDE is aimed at production agent teams who need to iterate quickly on agent logic without abandoning their standard development workflow. It's best described as "VS Code for agent systems."

Features

• MCP 2.2-native tooling — streaming tool results and cross-server chaining are visualized and editable from the IDE
• Skill Protocol v2 integration — write, sign, and publish agent skills directly from the IDE; enterprise registries supported
• Visual orchestration canvas — drag-and-drop agent topologies; same primitive set as LangGraph, but with live debugging
• Side-by-side trajectory inspection — compare two agent runs on the same input, diff the intermediate decisions
• Replay from any checkpoint — agents run in a sandboxed interpreter; you can pause, inspect state, and resume
• Built-in skill catalog — 800+ skills included with search, filter, and one-click install
• Runtime verification hooks — OpenShell-style property checks can be attached to agent steps and evaluated live
• Multi-provider backend — Claude, GPT, Gemini, DeepSeek, Mistral, and local Ollama models
• Distributed debugging — debug multi-node agent systems across machines with a single visual surface

Installation

# macOS
brew install --cask agenthub-ide

# Windows / Linux
# Download from https://agenthub-ide.dev/download

# Or via npx for the headless client
npx @agenthub-ide/cli init

Quick Start

// agent.ts — your agent logic
import { agent, tool } from "@agenthub-ide/sdk";
import { mcp } from "@agenthub-ide/mcp";

export default agent({
  name: "code-reviewer",
  steps: [
    tool("read_pr", { source: "github-mcp" }),
    tool("analyze", { model: "claude-opus-4-7" }),
    tool("summarize", { model: "claude-opus-4-7" }),
    tool("comment", { source: "github-mcp" }),
  ],
});

In the IDE, drag this agent onto the orchestration canvas, wire up MCP servers, and run. The IDE streams results, lets you pause at any step, and shows the full trajectory.

Advanced Features

• Hot reload — edit agent logic while a run is in flight; changes apply to the next step
• A/B run comparison — split the UI to run two versions of the same agent on the same input
• Skill protocol authoring — write skills inline, sign them, and push to a registry with one click
• Formal verification — attach property checks (invariants about what an agent can/cannot do) and evaluate them live
• Cross-server chaining visualizer — see the full MCP call graph including cross-server invocations

Examples

Debugging a flaky agent:

1. Run the agent on the flaky input
2. Open trajectory view — find where the run diverged from expected
3. Replay from the divergent checkpoint
4. Edit the agent logic
5. Re-run and diff against the original trajectory

Publishing a signed skill:

import { publish } from "@agenthub-ide/skill-protocol";

await publish({
  skill: "./skills/review-security.yaml",
  registry: "enterprise",
  signWith: "my-keypair.pem",
  version: "1.2.0",
});

Pros

• Built for agent development from day one, not retrofitted
• MCP 2.2 and Skill Protocol v2 native
• Visual debugging dramatically reduces agent iteration time
• Trajectory diff is genuinely useful for agent QA
• Multi-provider support without vendor lock-in

Cons

• Steeper learning curve than standard IDEs for first-time users
• New project — plugin ecosystem still forming
• macOS-first; Windows and Linux supported but with minor feature gaps

When to Use

• You're building production agent systems and want visual debugging
• You're already invested in MCP and want a coherent IDE around it
• You need to publish signed skills to enterprise registries
• You want trajectory diffing for agent QA

Resources

• Website: https://agenthub-ide.dev
• GitHub: https://github.com/agenthub-ide/client
• Docs: https://agenthub-ide.dev/docs