AgentBridge

The missing bridge between AI agents and the tools developers already use.

AgentBridge connects A2A agents (Google's Agent-to-Agent protocol) to MCP clients (Claude Code, Cursor, Copilot) — so your agents automatically appear as tools inside your IDE with zero per-agent configuration.

Claude Code / Cursor / Copilot
        ↕  MCP
  AgentBridge Registry  ←──── agents register themselves
        ↕  A2A
  code-review · test-writer · doc-generator · your-agent

---

Why AgentBridge?

Without AgentBridge	With AgentBridge
Configure each agent manually in every MCP client	Register once → available everywhere automatically
Agents can't talk to each other	Agents call each other via A2A protocol
No visibility into what agents are doing	Real-time dashboard: messages, health, latency
Testing agents requires terminal commands	Click → type → run from the dashboard UI

---

Quick Start

Prerequisites: Node 20+, an OpenRouter API key

git clone https://github.com/shrimaanshreyash/Agent-Bridge
cd Agent-Bridge
npm install
cp .env.example .env   # add your OPENROUTER_API_KEY
npm run build:dashboard
node packages/cli/dist/bin/agentbridge.js up

Open http://localhost:6100 — the dashboard is live with 3 agents running.

---

Build an Agent in 30 Seconds

import { BaseAgent } from '@agentbridge/core';

class SummarizerAgent extends BaseAgent {
  config = {
    name: 'summarizer',
    description: 'Summarizes any text in 3 bullet points',
    version: '1.0.0',
    capabilities: ['summarization', 'nlp'],
    inputs: { input: { type: 'string', required: true } },
    outputs: { summary: { type: 'string' } },
  };

  async execute(input: Record<string, unknown>) {
    const text = (input.input ?? input.text ?? '') as string;
    const summary = await this.callLLM(`Summarize in 3 bullets:\n\n${text}`, {
      system: 'You are a concise summarizer. Always return exactly 3 bullet points.',
    });
    return { summary };
  }
}

const agent = new SummarizerAgent();
agent.start(6104, 'http://localhost:6100');

Scaffold with the CLI:

node packages/cli/dist/bin/agentbridge.js init my-agent

---

Multi-Agent Workflows (YAML)

Define pipelines in agentbridge.yaml:

workflows:
  code-quality:
    description: "Review code → write tests → generate docs"
    steps:
      - agent: code-review
        input: $input
        output: $review
      - agent: test-writer
        input: $input
        output: $tests
        condition: $review.score < 90
      - agent: doc-generator
        input: $input
        output: $docs

---

Use Agents in Claude Code (MCP)

# Terminal 1 — start agents
node packages/cli/dist/bin/agentbridge.js up

# Terminal 2 — start MCP bridge
node packages/cli/dist/bin/agentbridge.js mcp

Add to your Claude Code MCP config:

{
  "mcpServers": {
    "agentbridge": {
      "command": "node",
      "args": ["packages/cli/dist/bin/agentbridge.js", "mcp"],
      "env": { "AGENTBRIDGE_REGISTRY": "http://localhost:6100" }
    }
  }
}

Every registered agent appears as a tool — automatically, no additional config.

---

Bring Your Own Framework

AgentBridge wraps existing agents without rewriting them:

import { LangChainAdapter } from '@agentbridge/adapter-langchain';

const adapter = new LangChainAdapter({
  name: 'my-langchain-agent',
  description: 'Existing LangChain agent',
  capabilities: ['research'],
  agent: myExistingChain, // RunnableSequence / AgentExecutor
});

await adapter.start(6105);
await adapter.register('http://localhost:6100');

Adapters available: LangChain, OpenAI Agents SDK, CrewAI

---

Dashboard

Open http://localhost:6100 while agentbridge up is running.

Page	What you see
Registry	All agents with status — click any to test it live from the browser
Messages	Real-time feed of every agent call with latency
Workflows	Interactive graph of your pipelines — drag nodes freely
Health	Success rates, response times (avg / p95 / p99) per agent

---

CLI Reference

agentbridge up                    # Start registry + all agents from agentbridge.yaml
agentbridge mcp                   # Start MCP bridge for Claude Code / Cursor
agentbridge dashboard             # Start dashboard standalone
agentbridge list                  # List registered agents
agentbridge call <name> <input>   # Invoke an agent from terminal
agentbridge init <name>           # Scaffold a new agent
agentbridge register [path]       # Manually register an agent

---

Project Structure

AgentBridge/
├── packages/
│   ├── core/                  — Registry, A2A protocol, BaseAgent, MCP bridge
│   ├── cli/                   — CLI commands
│   ├── dashboard/             — React dashboard (Vite + Tailwind + ReactFlow)
│   ├── adapter-langchain/
│   ├── adapter-openai-agents/
│   └── adapter-crewai/
└── agents/
    ├── code-review/           — Reviews code for bugs, security, best practices
    ├── test-writer/           — Generates unit tests (vitest / jest / mocha)
    └── doc-generator/         — Generates README, JSDoc, or API reference docs

---

Configuration

Variable	Description	Default
OPENROUTER_API_KEY	Required — get one free at openrouter.ai	—
MODEL	LLM model via OpenRouter	google/gemini-2.5-flash-lite

Recommended models: google/gemini-2.5-flash-lite (default, fast + cheap), anthropic/claude-3-haiku (reliable JSON), openai/gpt-4o-mini (strong reasoning).

---

License

MIT © 2026 Vemula Srimaan Shreyas