Books demo tour · Part 1 of 6
Your AI becomes your expert bookkeeper.
Two ways in, one rule underneath. Books now ships with its assistant built in: open the chat panel, ask the live ledger, hand it real work, and approve what posts - no setup at all. Or connect the AI your firm already licenses - Google Antigravity, Claude, ChatGPT, or Gemini Spark - over MCP. Either way it is the same scoped, recorded, revocable tool layer: AI prepares, and people approve the ledger.
Illustrative client book: Hudson & Vale CPAs, LLP keeping the books of Harborlight Community Health Center, Inc. - a fictional nonprofit community health center. Fictional firm, client, and data.
Books MCP server URL
https://book.workpaperos.com/api/book/mcpAdd this URL in your AI product's remote MCP settings and sign in with your WorkpaperOS account. Authorize each AI product separately; its one Books connection covers ledger tools and Files. File execution and network approvals depend on the AI client.
Built in - new
The Books Assistant is already inside the app
Open the chat panel in Books and the assistant is there - multiple chats with history, like ChatGPT or Claude: switch between them, start new ones, delete old ones. It answers from the live ledger - trial balance, balance sheet, P&L, AR and AP aging, account activity - real numbers, not summaries. Attach files to a message or pick from the Documents & Storage library; uploaded documents are indexed and OCR'd, and the assistant reads their contents. And it does the work: journal entries, bank-feed categorization and matching, bank reconciliations, invoice and bill drafts - every posting stopping at an in-chat approval card first.
Approval required · journal entry draft
Dr Building & occupancy — $1,850.00
Cr 1010 Operating Cash — $1,850.00
Memo: Meridian Facilities - June janitorial · support: meridian-june.pdf
Every posting stops at a card like that one - the approval is the click, and it is yours. The categorizations queue their own cards the same way.
The bot does the heavy lifting
- Answers from the live ledger - trial balance, balance sheet, P&L, AR/AP aging, account activity - with the actual figures.
- Reads your documents: attach files to a message or pick from the Documents & Storage library; uploads are indexed and OCR'd.
- Does the work: creates journal entries, categorizes and matches bank-feed transactions, runs bank reconciliations, drafts invoices and bills.
You stay in control
- Every posting stops at an in-chat approval card first - nothing reaches the ledger silently.
- You pick the model per chat: Gemini, Claude, GPT, or an open model. Every plan can use every model.
- Every action runs through the same audited, firm-scoped tool layer as the MCP connector.
Models and plans
Credits weight by model class - open 0.5, fast 1, standard 3, max 8 per message - with live weekly and 5-hour meters in the app. Near the limit, requests fall back to the fast default instead of stopping.
One URL
Add the Books connector where your AI lives
Prefer the assistant your firm already licenses? Connecting your own AI over MCP remains part of your app plan - same tools, same approval gates, same audit trail. Paste the Books MCP URL into your assistant's connector settings. A secure sign-in does the rest, and access is scoped to exactly the companies you authorize.
Books MCP server URL
https://book.workpaperos.com/api/book/mcpUse your AI product's remote MCP/custom-connector settings. Complete OAuth sign-in and select the client companies it may access; no second Files connector is required.
The bot does the heavy lifting
- Sees the full Books toolset after authorization: bank feeds and statement imports, categorization, invoices and bills, journals, reconciliation, close, reports - and the United States payroll module, which is the only payroll module there is today.
- Every write is a proposal - the tools are literally named propose_categorization, propose_match, propose_pay_bill.
- Works from Antigravity, Claude, ChatGPT, or Gemini Spark on your AI subscription.
You stay in control
- You choose which client companies each connector may touch.
- Proposals post to the books only when a person approves them.
- Every authorized tool call is recorded in the audit trail; revoke access any time.
How they fit together
Built in, or your own AI over MCP
The built-in assistant needs no setup. The connector puts the same tools inside whatever AI your firm licenses. Pick one, or run both - either way it is the same scoped, recorded tool layer.
| Route | What the AI works with | Setup |
|---|---|---|
| Built-in Books Assistant | The live ledger and your indexed documents, in the app's own chat panel. | none - it is already there |
| Your AI over MCP | The live ledger, indexed document search, and original-file exchange with a permitted AI analysis environment - using the same scoped authorization and audit trail. | paste one URL |
Books Files
Original files to your AI. Finished work back to your file room.
Connect Books once in each AI product and authorize the companies it may access. The same MCP connection covers records, indexed document search, and file exchange: no second Files connector or desktop folder mirror. Your assistant can request the original Excel, PDF, Word, or image file, analyze it in a supported execution environment, and upload a generated file to the authorized client book.
Works with ChatGPT, Claude, and Google Antigravity
The original-file download, analysis, and return-upload workflow has been successfully validated with all three. Claude and Google Antigravity have completed manual validation alongside ChatGPT. Use your existing product connection; no second Files connector is needed.
1 · Authorize
Sign in with your WorkpaperOS account through the product's MCP connector. Access remains company-scoped and revocable.
2 · Analyze originals
MCP obtains a short-lived download link. Original bytes travel over HTTPS outside the conversation's tool payload, with file identity and integrity information.
3 · Return the work
The assistant reserves an upload, transfers the file over HTTPS, and completes validation. Supported large-file uploads can resume; finalized files enter the app's indexing pipeline.
| Your AI | How file exchange runs |
|---|---|
| Gemini Spark | Connect through Custom apps for Spark using the same product MCP URL. Indexed data and tools are available through the authorized connection; original-file download, analysis, and return upload depend on Spark's available file-execution and HTTPS capabilities. |
| ChatGPT | Uses the supported attachment handoff or an authorized HTTPS transfer in its file-analysis environment. File access may require approval in the current chat. |
| Claude | Uses the shared transfer tools. Claude Code can transfer files from its permitted workspace; Claude web/Desktop chat requires an available file-execution environment and approved network access. |
| Google Antigravity | Uses the same remote MCP connection and HTTPS transfers from its permitted agent workspace. Terminal, file, and network permissions still apply. |
Connect Gemini Spark
Use Gemini Spark with Books through the same MCP connection approach as ChatGPT and Claude. Ask for work on your authorized company; review the results and approve writes when prompted.
- In Gemini web, open Settings & help → Connected Apps → Custom apps for Spark → Add a custom app.
- Paste
, then follow the sign-in and authorization prompts for your WorkpaperOS account. - In Spark, type @ and select your connected app, then describe the task, company, and period.
Google currently requires Spark access, a personal Google Account, age 18+, US availability, English, and Keep Activity enabled. Check your firm's confidentiality policy and Google's data settings before connecting client information. Custom apps are for Spark, not ordinary Gemini chat. Original-file analysis and return uploads depend on Spark's available file and network capabilities.
Google's current Spark setup and eligibility requirementsProjects
Give your bookkeeper a Project
Use a supported project or workspace in your chosen AI product for each client book. Load the firm's templates, categorization rules, checklists, and appropriate prior-year context. Project copies are snapshots: use the Books connector for live ledger data, indexed search, and current original files. Return completed schedules and supporting documents through Books Files rather than maintaining a separate folder mirror.
| Project knowledge | What the AI does with it |
|---|---|
| 📄 Firm templates & categorization rules | studied - categorizes the way your firm does |
| 📄 Harborlight FY2025 closed books (prior year) | studied - knows the account patterns |
| 📁 Client folder snapshot - receipts, bills, statements | uploaded copies - refresh them when they change |
Connector: WorkpaperOS Books (MCP) · every chat starts already knowing the client and your conventions. For live numbers rather than a snapshot, the connector hands the same chat the ledger and your indexed documents.
First conversation
Talk to it like you'd talk to your bookkeeper
▶ Watch: connecting your AI to WorkpaperOS Books
Screens are illustrative: Hudson & Vale CPAs and Harborlight Community Health Center, Inc. are fictional.