Skip to main content
The Zeno operator gives Claude your hands on the app. It signs in as you, reaches every module your own login reaches, and writes through the same API your browser does — so posting, dedup, and the audit stamp all stay server-side. Connecting is a one-time OAuth login. There are no API keys to copy.

​
Connect in Claude Code

That’s the whole install. The connection carries its own instructions — see below.

​
Connect in Claude Desktop

Open Settings → Connectors, add a custom connector, and give it this URL:

​
Signing in

The first time you use the connection, Claude opens a browser at api.zenorent.com. Log in as yourself and approve. Claude stores the token; you never see it. The token is bound to the organization you approved from — if you belong to several, switch organizations in Zeno before you approve, or approve once per organization.
Approve only your own login. The connection carries exactly the access your account has, so approving from an admin account hands Claude admin reach.

​
Where Claude’s instructions come from

Nothing is installed on your machine, and there’s no prompt to paste. Everything Claude needs arrives over the connection:
  • How to work in Zeno. The server sends operating instructions when Claude connects: how to discover capabilities, follow operation results, and distinguish a reference number from a database id.
  • A field guide. A built-in reference maps the common jobs to exact endpoint sequences, so Claude follows the worked path instead of rediscovering it.
  • A map of your API. One row per module with its endpoint count, so Claude orients without pulling ~1,700 endpoints into the conversation.
  • Skills for recurring work. Short descriptions are searchable; full instructions load when needed. See below.
Because these ship with the server, everyone in your organization gets the same behavior, and improvements reach you without you updating anything.

​
Tailored to your work

Your AI asks what you do and what you want help with, one useful question at a time. You can choose a starting point such as dispatcher, bookkeeper, technician, office manager, or owner/manager, or describe your own responsibilities. It remembers your answers separately for each organization. These preferences shape questions and suggestions. They do not change your permissions or hide authorized tools. You can change them by telling your AI. Setup is optional and does not block your current task.

​
Your own slash-commands

A Skill saves instructions for recurring work: processing an insurance referral, checking a job before dispatch, recording a remittance, or closing the month. Every user can use Skills, create personal Skills, and propose team Skills. Your AI introduces the feature with an example relevant to your work. Your AI searches names and descriptions first, then loads the relevant instructions. Active Skills also appear as slash-commands (/<slug>) in clients that support MCP prompts. These instructions are served by Zeno; they are not installed as native client Skill packages. To create one:
  1. Work through the task with your AI. Ask it to save the recurring steps as a Skill. It helps fill in missing decisions, inputs, exceptions, and completion checks.
  2. Choose who can use it. Personal Skills stay private. Team drafts are visible to their author and work-area managers; active team Skills are available to users with access to that work area.
  3. Review and activate. Follow the returned review link or open Skills in the sidebar. You can activate your own personal Skill; a work-area manager activates a team Skill. The AI cannot activate its own draft.
The Skills page also lets you create and edit instructions directly, declare inputs, and move a Skill to the right work area. For example, an insurance intake procedure filed under Books can be moved to Glass without changing its command name. A work-area manager can propose sharing an existing procedure with the team, even if its original author is no longer available. Existing Books commands remain readable by users of the enabled Books agent; other existing procedures remain visible to work-area managers. Saving an edit returns the Skill to draft. Activation applies to the version you reviewed; a concurrent change requires you to reload. Skills never grant additional permissions or replace fresh business data. Use Disable Skill to pause a reviewed Skill without losing its approval. Enable Skill restores it without editing its instructions. Approving a draft that you saved disabled keeps it disabled until you enable it.
A fact about a vendor belongs in org knowledge; a pattern that always maps to the same vendor and GL account belongs in a rule. A Skill captures a reusable workflow. See Rules.

​
Check that it worked

Ask Claude which organization it’s in. It answers from your live profile, so a wrong answer means you approved from the wrong organization — revoke and reconnect. Then ask what it can reach. It replies with the module map — one row per module with an endpoint count — rather than a wall of endpoints.

​
Using a static token instead

For the book CLI, a script, or raw HTTP, ask us to mint you an operator token. Store it in ~/.zenorc or export it as ZENO_BOOKKEEPING_TOKEN, then call any /api/... path with an Authorization: Bearer … header. Review expiry and revoke from Settings → Agent Tokens; revoking takes effect immediately. It’s the same token kind either way — the OAuth flow just mints and stores one for you.

​
What it can and can’t do

  • It’s you, bounded by you. The connection grants exactly what your own login grants: your role, your organization. It never elevates. If you can’t do something in the app, Claude can’t either.
  • It follows the server’s review rules. Routine authorized work runs through the API. If an operation returns a prepared state, Claude links you to its review page; preparation is not completion. The current effect classification determines when review is required.
  • Documents go through a review gate. Turning a PDF into a bill runs through the server’s parse-preview, and the server compares its own confidence against Claude’s read. Disagreement or low confidence blocks the entry until you approve it. See Bulk bookkeeping.
  • Everything is audited under you. Writes carry your user and the session id, so the audit trail shows what was done, when, and through which session.

​
Where to go next