Skip to content
Eternify Labs

MCP is the foundation

Your books, in any agent.

Eternify Ledger runs an MCP server, so Claude, ChatGPT, Microsoft Copilot Studio, Cursor or your own agent can read your books and draft the work, with the permissions of the person who connected it. A named person approves every change.

MCP connector, in motion
  • ReceivablesAgeing, late invoices and reminders drafted.
  • PayablesBills drafted from a PDF, payment runs proposed.
  • TaxThe GST/HST figure and the return, built from posted entries.
  • ReconciliationOpen bank lines, with matches proposed.

What you get

Any MCP client

Claude custom connectors, ChatGPT in developer mode, Microsoft Copilot Studio, Cursor, or an agent your team wrote. One connector, no plug-in per assistant.

The person's role, nothing more

The connection signs in as a person. The agent has their companies, their role and their limits, so it can never see or do more than they can.

Reads answer, writes wait

Reading tools answer from posted entries. Writing tools draft into the review queue, and nothing posts until a named person approves it.

How it works

Five steps, and the approval is yours.

Pick a step to see it. The panel stays beside you while you read.

Add Eternify Ledger as a connector in your assistant and sign in as yourself. The assistant gets your role in the ledger and nothing beyond it, and you can remove the connection at any time.

  • Sign in as yourself, with your existing role
  • Scoped to the companies you may open
  • Revoked in one click, with the revocation logged

Adding the connector

  1. Add the connectorIn Claude, ChatGPT or Copilot StudioInfo
  2. Sign in as yourselfYour role and your companiesInfo
  3. ConnectedRead tools answer, write tools waitDone

Type it the way you would ask a colleague. Answers come from posted entries, with every figure traceable to the record behind it, so a number in a chat can be checked in the ledger.

  • Receivables, tax, reconciliation, budgets and reports
  • Figures link to the entries they came from
  • Nothing is guessed from outside the books

What do we owe for GST/HST this quarter, and when is it due?

eternify.tax.return.gst_hst ok

eternify.compliance.calendar ok

18,904.12 for the quarter, built from posted entries. It is due on the date on the compliance calendar, and the return is ready to review.

Drafted bills, entries and payment runs land in the review queue with their source attached. The agent's name and the approver's name are both on the record, and nothing posts until a person signs off.

  • Drafts arrive with the PDF or the source attached
  • Routed to whoever approves that amount
  • Both names stay in the audit log

Review queue

Northline Supply, NS-2210Drafted by your agent, PDF attached4,812.00Awaiting approver
Payment run, 14 billsProposed by your agent61,204.18Awaiting approver
Office rent, OctoberApproved by the controller6,400.00Posted

A firm can run its own model against our queue. The agent leases a piece of work, does it with the firm's own inference, and submits a proposal. We run no inference for that client, and the client's data never touches our model provider.

  • Work leased from a queue, one item at a time
  • The firm's model, the firm's credentials
  • Proposals reviewed like any other

A leased work item

  1. LeaseYour agent takes one item from the queueInfo
  2. WorkYour model, on your sideInfo
  3. SubmitA proposal enters the review queueWaiting
  4. ApprovedA named person signs offDone

An agent's tools are split in two. Read tools answer straight away, within the user's role. Write tools draft, and the draft waits in the queue for the person whose authority covers it. A refusal comes back as a plain result the agent can read, so the connection stays up.

  • Read tools answer
  • Write tools wait for approval
  • Errors come back as results, never a dropped connection

Some of the tools your agent sees

ToolModeState
receivables.ageingReadAnswers
tax.return.gst_hstReadAnswers
reconciliation.open_linesReadAnswers
budgets.varianceReadAnswers
bills.draftWriteNeeds approval
payments.propose_runWriteNeeds approval
orders.draftWriteNeeds approval

Adding the connector

  1. Add the connectorIn Claude, ChatGPT or Copilot StudioInfo
  2. Sign in as yourselfYour role and your companiesInfo
  3. ConnectedRead tools answer, write tools waitDone

Architecture

Your agent never posts.

The connector is a front door to the same ledger. Every call is checked against the signed-in person's role, and every write is a proposal.

Capabilities

What your agent can do

  • Receivables

    Ageing, late invoices and reminders drafted.

  • Payables

    Bills drafted from a PDF, payment runs proposed.

  • Tax

    The GST/HST figure and the return, built from posted entries.

  • Reconciliation

    Open bank lines, with matches proposed.

  • Budgets

    Budget, actual and the difference, by account.

  • Purchasing

    Purchase orders drafted and checked against the budget.

Keep going with the rest.

  • Eternify Assistant

    Assistant, agents and continuous work, with a named person approving every change.

    Learn more
  • Controls and approvals

    The agent proposes, people approve.

    Learn more
  • API and webhooks

    Build on your books.

    Learn more

Questions

Questions, answered.

What is an MCP server?

MCP, the Model Context Protocol, is an open standard that lets an AI assistant connect to outside software through one common interface. Anthropic introduced it in November 2024. An MCP server is the piece a product runs so that any MCP client, such as Claude or ChatGPT, can use it as a tool.

Can ChatGPT or Claude connect to accounting software?

Yes, when the accounting software runs an MCP server. Claude accepts custom connectors, ChatGPT accepts remote MCP servers in developer mode on its business plans, and Microsoft Copilot Studio adds MCP servers to its agents. The Eternify Ledger connector works with each of them and with any other MCP client.

Is it safe to let an AI agent change accounting records?

It is when the agent cannot post. Through the connector an agent reads what its user may read and drafts what its user may draft. Every change arrives in the review queue as a proposal with its source, and only a named person can approve it. The audit log records both.

Does the agent see more than the person who connected it?

No. The connection signs in as that person, so the agent has their role, their companies and their limits. A bookkeeper's agent can draft a bill, and it cannot release a payment run that needs an approver.

Can our accounting firm use its own AI model?

Yes. A firm's agent leases work from a queue and submits proposals, using the firm's own model and credentials. We run no inference for that client, and the client's data does not reach our model provider.

Connect your agent.

We connect your assistant to a working ledger on the call, and you ask it the questions your month-end asks.