Skip to main content
toolbox

Your team's toolboxfor AI agents.

Build skills, connect services, and write scripts once, in a library your whole team shares. Everyone puts together toolboxes from it and connects their own AI app. Credentials stay on your server, and risky calls wait for someone to approve them.

Create accountSee what's inside

Members can create an account with an approved work email or join by invitation.

Works with your coding agent

  • Claude Code
  • Codex
  • Cursor
  • Amp
  • OpenCode
  • GitHub Copilot
  • Gemini CLI
  • Windsurf
  • + any MCP client
ToolboxesCustomer supportExample

Connections

  • LinearRemote MCP

    4 of 12 tools enabled

    Ready

  • StripeAPI

    3 of 40 tools enabled

    ReadyRefunds ask first

Skills and resources

  • Reply to a refund requestSkill

    Team library, revision 7

    Current

  • Refund policyResource

    PDF, revision 3

    Current

Scripts and inbox

  • weekly_refund_reportScript

    Uses Stripe and Linear

    Reviewed

  • [email protected]Email

    2 new messages

    Receiving

  • Stripe eventsWebhook

    Sender verified by script

    Receiving
Connect an agentdiscover · call · execute
https://toolbox.acme.dev/mcp/aaron/customer-support

Linear and Stripe sign-ins stay on your server.

What goes in a toolbox

A toolbox points to items in your library. Nothing is copied, so one skill, connection, or script can belong to several toolboxes.

  • Skills and resources.

    Skills teach agents how your team works. Resources give them reference material such as policies, PDFs, images, and notes.

    • Every save is a revision you can compare
    • Keep them personal or share with the team

  • Connections.

    Add a remote MCP server or import an OpenAPI or Google Discovery description, then sign in to the account agents should use.

    • Work and personal accounts, side by side
    • New operations stay off until you opt in

  • Scripts.

    Agents write short JavaScript that chains operations, so a ten-step job is one request instead of ten. Save the good ones to reuse.

    • Sandboxed to the operations you enabled
    • Team scripts can require review first

  • Email and webhooks.

    Each toolbox gets its own email address and webhook receivers. Agents read and acknowledge what arrived when you ask them to check.

    • Validation scripts verify webhook senders
    • No mailbox or signing secret reaches agents

Agents do the work. You keep control.

toolbox runs on your own server, so your team decides what agents can reach and who signs off.

Credentials stay home

Sign-ins and API keys never reach the AI client. It gets a toolbox URL and nothing else.

  • Approval for risky calls

    Set operations like refunds or deletes to wait for a person to approve each call.

  • Team review

    Admins review shared skills, resources, and scripts, and decide who gets team integrations.

  • Audit log

    Changes to toolboxes, connections, access, and team content are recorded with who made them.

Create an account, then verify your work email to continue.

Then paste a toolbox URL into any AI app that supports MCP. Connect with a sign-in flow or a separate token for each app.

Create account