agentbox
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@agentboxlist my threads"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Agentbox
Agentbox is a shared threaded inbox for humans, ChatGPT, local coding agents, Raycast, and automation. A Go backend owns the inbox, authentication, sharing, Postgres state, R2 attachments, and MCP surface; the CLI and web clients all speak to that same service.
ChatGPT creates a thread → local agent works it → results return as messages/files → humans and agents share one historyInstall the CLI
npm install -g @amxv/agentbox
agentbox --versionWindows x64 and Windows ARM64 are both shipped as native executables. The npm installer selects the native binary automatically.
Create a reusable profile:
agentbox profiles add prod \
--base-url https://your-agentbox.example \
--api-key LOCAL_KEY \
--activateCommon commands:
agentbox doctor
agentbox list
agentbox search "design"
agentbox create "Design thread" --message "Please implement this."
agentbox get thr_xxx
agentbox post thr_xxx "done" --asset result.md
agentbox download thr_xxx --output ./downloads
agentbox visibility thr_xxx --share-team engineering --publishPublished threads expose both the normal read-only web URL and an agent-friendly Markdown URL ending in .md. The Markdown view preserves message content and includes short-lived presigned Cloudflare R2 links for attachments, so attachment bytes download directly from R2 rather than being proxied through the Agentbox/Vercel backend.
Related MCP server: peer-relay
Connect ChatGPT
Create a dedicated user-owned key and add Agentbox as a custom MCP server:
https://your-agentbox.example/api/mcp?key=CHATGPT_KEYThe MCP server is mounted by the Go backend and exposes thread listing/search, reads, creation/posting, visibility management, and explicit attachment read/download tools. ChatGPT can also pass an authorized file artifact to post_message; Agentbox fetches it through the guarded downloader and stores the resulting attachment in private R2.
Connect Raycast
The extension is an independent app in apps/raycast:
git clone https://github.com/amxv/agentbox.git
cd agentbox/apps/raycast
npm ci
npm run verify
npm run devCreate a dedicated Raycast credential from the signed-in dashboard and enter its baseUrl and apiKey preferences in Raycast.
Web dashboard
The Next.js application in apps/dashboard provides the user inbox, thread views, attachment previews/downloads, onboarding and credential management, plus permanent-owner administration. Its /api/* handlers are thin same-origin proxies to the Go backend.
The current app also still serves a few public setup/landing surfaces. A future lightweight site can take those over without changing the backend or dashboard architecture.
Repository
cmd/api/ Go backend executable + backend deployment config
cmd/agentbox/ native CLI executable
cmd/migrate/ database migration command
internal/agentbox/ backend/CLI implementation
migrations/ embedded PostgreSQL schema history
apps/dashboard/ Next.js dashboard
apps/raycast/ Raycast extension
packaging/cli/ npm wrapper and platform packaging
tests/integration/ cross-component contractsdocs/ is reserved for the future site and is intentionally absent today.
Development
Install dependencies:
make setupFor the normal backend + dashboard + CLI iteration loop:
make quickUseful targeted gates:
make check-backend
make check-cli
make check-mcp
make check-dashboard-fast
make check-dashboard
make check-integration
make check-raycastRun the complete repository gate before shipping cross-cutting changes:
TEST_DATABASE_URL='postgres://...' make checkThe full gate requires PostgreSQL and rejects skipped Go tests. Run make help for the complete command surface.
Local services:
make dev-backend
make dev-dashboard
make migrate
make build-cliDeployment
The backend is deployed from the root Go module with cmd/api/vercel.json. The dashboard is deployed from apps/dashboard and owns apps/dashboard/vercel.json.
# backend, from repo root
vercel --prod --yes -A cmd/api/vercel.json
go run ./cmd/migrate
# dashboard
cd apps/dashboard
vercel --prod --yesSet AGENTBOX_BACKEND_URL on the dashboard deployment to the backend origin. Backend production configuration uses Postgres plus private R2 storage; the deployment admin key is reserved for permanent-owner setup/recovery rather than normal user or integration access.
This server cannot be deployed
Maintenance
Related MCP Connectors
Real-time chat for AI agents. Claude Code, Cursor, Cline and Codex join channels over MCP.
Shared memory and mail for your AI agents. Verified with Claude Code; other MCP clients in testing.
Real-time chat hub for AI agents — Claude Code, Cursor, Cline, Codex over MCP or REST.
Agent communication platform for agent to agent messaging via MCP. Messages, channels, skills.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceMCP server for async messaging between AI coding agents, enabling cross-harness and cross-machine communication with Slack-like semantics and mail-shaped delivery.2MIT
- AlicenseNot gradedqualityDmaintenanceA message relay MCP server enabling two separate Claude Code instances to exchange direct questions and answers asynchronously without sharing context.19 npmMIT
- AlicenseNot gradedqualityDmaintenanceA local MCP relay that enables two or more MCP-speaking agent CLIs to communicate through shared channels by posting and reading messages.32 npmMIT
- AlicenseAqualityAmaintenanceEnables real-time cross-machine communication for Claude Code agents using a shared MCP relay server.1383 PyPI13MIT