Connect ChatGPT and Claude to your business workspace

Most teams already have AI assistants in chat. Few of them are connected to the actual business: deals, tickets, CMS content, follow-up tasks. This post shows how to wire ChatGPT or Claude into a Slab5 workspace through MCP, with scopes you can defend.

What MCP is

MCP — Model Context Protocol — is an open standard for exposing tools to AI agents. An MCP server publishes a typed catalog of tools. An MCP client (a Claude or ChatGPT integration) calls those tools on the user's behalf.

How Slab5 fits

Slab5 publishes one MCP server per workspace at mcp.slab5.com/v1. The tool catalog mirrors the REST API — create_contact, search_contacts, create_task, update_support_ticket, create_entry, run_bi_query, and so on. Each tool call checks the credential's scopes against the workspace's module permissions.

Connect with a static MCP client (personal use)

In the dashboard, create an MCP client for your workspace and grant it the scopes you need: crm:read, tasks:write, support:read, and so on. Copy the token (returned once) and paste it into Claude or ChatGPT as the MCP server credential.

Connect with OAuth (org-wide rollouts)

For organizations, create an OAuth-capable MCP client. Each team member authorizes a subject grant against the same workspace. Revoking a grant revokes that user's agent access without affecting the rest of the org.

Defensive defaults

Start read-only. Add write scopes one module at a time. Use the audit log to confirm the agent's behavior matches your intent before broadening scopes.

Add MCP client with chosen scopes. Agent issues tool calls that map 1:1 to REST endpoints. Tool catalog is auto-scoped to the client's permissions.

Server automations continue to use API keys. REST and MCP share the same workspace data, scopes, audit log, and usage metering.