AgentBridge

A fast, simple Kanban board where humans and AI agents collaborate on tasks. AgentBridge runs on Cloudflare Workers and speaks the Model Context Protocol (MCP), so Claude, Cursor, Codex, and your team can work from the same source of truth.

Live demo: https://agent-bridge.0xkaz.com/

Features

• MCP server over HTTP/SSE (/mcp)

• REST API with revocable tokens

• Real-time board updates via SSE

• Google OAuth sign-in

• Personal org per user (no accidental domain sharing)

• Multi-project Kanban boards

• Custom lists (columns) per project

• Drag-and-drop task movement

• Task comments for agent memory and context

• Organization member invitations

Related MCP server: Kanban MCP Plugin

MCP Tools

• list_orgs — list organizations

• list_projects — list projects in an organization

• create_task — create a task in a project

• update_status — change a task status

• comment — add a comment to a task

REST API

Most endpoints require Authorization: Bearer <token>. Invitation info and accept endpoints are public.

Setup

npm install
cp .env.example .env
cp wrangler.example.toml wrangler.toml
# Edit .env with your Google OAuth client ID
# Edit wrangler.toml with your Cloudflare account_id, D1 database_id, and custom domain
npm run db:seed

Google Cloud Console configuration

This app uses Google Identity Services (ID token flow). You do not need to configure a redirect_uri.

1. Create an OAuth 2.0 Web application client ID at https://console.cloud.google.com/apis/credentials

2. Add the following Authorized JavaScript origins:

   • http://localhost:8787 (local wrangler dev)

   • https://<your-worker>.workers.dev (production)

3. Copy the Client ID into .env as VITE_GOOGLE_CLIENT_ID and into .dev.vars as GOOGLE_CLIENT_ID

4. Copy wrangler.example.toml to wrangler.toml and fill in your Cloudflare account_id, D1 database_id, and custom domain

The Client Secret is not used by this app because verification is done with Google's public JWKS.

Development

npm run dev

Open http://localhost:8787 and sign in with Google.

For MCP clients, configure the endpoint:

{
  "mcpServers": {
    "agentbridge": {
      "url": "http://localhost:8787/mcp"
    }
  }
}

Generate an API token on the Settings page and use it as Authorization: Bearer <token>.

End-to-end tests

make e2e

This starts the dev server automatically and runs Playwright against Chromium.

Migrating an existing database

src/db/schema.sql only contains the final schema for fresh databases. If you are updating an existing D1 database from a version that stored plaintext token values, run the token-hash migration before deploying:

# Local
make db-migrate-token-hash

# Remote (production)
make db-migrate-token-hash-remote

These commands hash any remaining plaintext tokens, then recreate the tokens table with token_hash only.

Deploy

npm run deploy

Before deploying, make sure wrangler.toml is filled in (see Setup step 4).

Set secrets:

wrangler secret put GOOGLE_CLIENT_ID

Commands

• make lint — TypeScript type check

• make test — run tests

• make build — build frontend

• make db-seed — apply D1 schema locally

• make db-migrate-token-hash — migrate local D1 tokens to hashed schema

• make db-migrate-token-hash-remote — migrate remote D1 tokens to hashed schema

• make e2e — run Playwright end-to-end tests

• make dev — start dev server

License

MIT