Agent access
Your AI can work in your workspace.
Every workspace can serve a Model Context Protocol (MCP) endpoint at your‑org.everjust.app/mcp. Point Claude Code, Codex, or any MCP client at it and your AI reads and updates the same CRM, invoices, and records your team uses — with your permission, never past it.
> which customers haven't been invoiced this month?
calling search('crm.lead', …) · search('account.move', …)
Three customers with won deals and no June invoice:
Meridian Supply, Oak & Iron Co., Fieldstone LLC.
> draft the invoice for Meridian based on their quote
calling create('account.move', …) — logged to everjust.mcp.log
Draft invoice INV/2026/0142 created from quote SO0087.
It's in Invoicing for your review — nothing sends without you.Connect in two steps
- Mint a key. In your workspace: Preferences → Connect an AI agent. One click creates a scoped API key and shows your config.
- Paste the config. For Claude Code:
claude mcp add --transport http everjust \
https://your-org.everjust.app/mcp \
--header "Authorization: Bearer <YOUR_API_KEY>"Any MCP-compatible client works the same way — the endpoint speaks standard JSON-RPC over HTTP with your workspace API key as the bearer token.
What the agent can do
Search, read, create, and update any record its user is allowed to touch, and discover your data model as it works — ten tools covering the whole platform surface.
The security model
Bounded by design, logged by default.
Your key, your rules
The agent runs as the user who minted the key — it can only see and change what that person can. There is no separate agent super-user.
Destructive actions gated
Deletes and writes require an explicit confirmation step. Even then, the action only goes through if the key's user is allowed to do it — confirming doesn't grant new rights.
Every action logged
Each call writes an audit row — who, what, which records, success or error — so you can review everything your AI did.
Agents aren't seats
Connected AI agents never count toward your 12 people. Connect as many as your workflows need.
Plain words
What we do and don’t sell.
Agent licenses give your AI tools working access to your workspace, and we set up your first agent workflows with you. We don’t sell autonomous "AI employees" — your agents act with your permission, under your access rights, on the record. When that changes, this page will say so.
Which AI tools work?
Anything that speaks MCP over HTTP: Claude Code, Codex (via mcp-remote), and other MCP-compatible clients. If your tool can send a bearer token, it can connect.
Do we need the With Agents license?
Agent access ships with the $2,500/month With Agents license and is included in the New Business program. It covers the endpoint, one-click connect for your team, and we set up your first agent workflows with you. See Pricing.
Can the agent leak our data?
Two different questions. Inside the platform: the agent sees only what its key’s user can see, on your own isolated database, keys can be revoked at any time, and the audit log shows every call. Outside it: your AI runs on the provider you chose — Claude, Codex, whoever — so anything the agent reads is processed by that provider under your agreement with them. The platform itself sends your data to no AI vendor. If there are records your agent should never read, scope the key’s user so it can’t.
Next step
See an agent work a live workspace.
Bring your own AI to the walkthrough — we’ll connect it live.