what it does, and what it won't
A lead AI agent and seven specialists, for the work around your notices and letters. They read first and draft second. You check, decide and send.
It all runs inside your own Claude. No website, hosting or code to set up. The full system, at the end, is your own setup around the kit, in the order to set it up.
what it doesRead the property's history in PropertyMe and its folder, and the whole email thread in your inbox, then draft a calm, factual reply that answers every point raised, with [MISSING] wherever it can't confirm a fact, and a table of any legal point: the Act, the section or clause, and the date. Then it checks the draft against your rules and reads it the way the tenant or owner will. If it's a complaint to a regulator, or there's a lawyer or the tribunal in it, it's built to flag it, not treat it as routine. It comes back as text in the chat, for you to paste into a reply in Outlook or Gmail
what the kit tells it never to doSend it, or put a draft in your mailbox. Promise a refund, a repair date or a bond outcome you haven't decided, admit fault, or make up a fact to fill a gap
what it doesWrite one log entry for every finished job, in your name: the date, who did it, what was done and why, who approved it, and where it's filed. You make the log yourself, once, in the property's top folder: a spreadsheet or list with those five columns, or a document. Claude writes each entry; check it, then say OK. Where your connector can add a row to your list or spreadsheet, it adds it; otherwise you paste it in. It also writes one short line for your CRM note, for you to paste into PropertyMe. Why both? PropertyMe says its connector is read only, so Claude can't add the note there, and the log is what Claude reads back before the next job, where your connector reaches it (otherwise paste the last few entries in). Did something without Claude, like a phone call? Type log this:, then one line saying what you did
what the kit tells it never to doCreate the log file for you (the kit is built never to save a file), save anything before you say OK, edit or delete an old entry (a mistake gets a correction entry), or write in PropertyMe
what it doesTake the entry condition report and the exit report, or the last routine inspection report, from your folder or as files you attach, and set them side by side, room by room, item by item, in each report's own words, marking what's changed. You get a table for each room, a numbered list of the differences and the gaps, as text to paste into Word or Google Docs
what the kit tells it never to doDecide what's fair wear and tear, or what goes in a claim: that's your call. It compares words only, never photos, and never pulls the reports out of PropertyMe
what it doesAsk in plain words, about one property or your whole rent roll: the leases coming up for renewal, the inspections booked or the open jobs, with the properties, contacts and tasks behind them. If your lists live in Airtable, it can read those too. You get a table that says when it was read, with the record behind every row, and what it couldn't include
what the kit tells it never to doPull or work out arrears, rent, ledgers or bonds. Give those yourself, from your trust ledger, with the date. Any other report: ask it, and check what your connector can see before you rely on it
what it doesRead PropertyMe, Microsoft 365 or Google, and Airtable if you connect it, with your own login, so it only sees what you can already see. It asks Allow once before each read
what the kit tells it never to doChange anything in PropertyMe (PropertyMe says its connector is read only), create or change a record in Airtable, send an email, or save, move, rename or delete a file. The only thing it adds is one new log entry, after you say OK
Every legal point should name the Act, the section, clause or regulation, and the date. If it doesn't, ask before you rely on it. Tenancy law differs by state, and the kit checks your own state's law. PropertyMe itself says to check key figures before you act on them, and to ask for the records behind any answer. This kit is not legal advice.
what you need
Eight things, and you probably have most of them. Airtable is only for agencies whose lists live there.
Pro or Max, from claude.ai (Team and Enterprise work too). The kit's lead is a skill (a written procedure Claude follows), and Anthropic says skills are available on Pro, Max, Team and Enterprise plans. Keep client work on the Claude account on your work email, never one on your personal email. You use it in chat, on claude.ai or the Claude desktop app. A difficult email is a bigger job than a normal chat, so if you run a lot of them, you may need a higher plan.
set up by: YouThe Claude setting that lets Claude run small tools and make files inside a chat; skills switch on through it, and Anthropic says skills need it. The kit itself holds no code and builds no file. It's on by default; to check, select your name, then Settings, then Capabilities.
set up by: YouThe Claude setting that lets it read websites. Here it reads your state's tenancy law on the official legislation website. In a chat, select + at the bottom left of the message box, then Web search. No Web search there? Your Claude searches the web on its own.
set up by: YouA connector is a link that lets Claude open one of your work apps with your own login, so it only sees what you can already see. AiMe Connect comes with PropertyMe's Advanced plan and above, and PropertyMe says it's read only: Claude can look things up, but it can't change anything in PropertyMe. PropertyMe says it can read properties, contacts, jobs, tasks, inspections, lease renewals and team members. Your agency chooses who gets it. Not on PropertyMe's Advanced plan? You can still use the kit: skip the PropertyMe steps and paste the records into your first message instead (the property's details and the email for a reply, the two reports for a side by side, your list for a report). It works from what you give it. On another CRM, use its own Claude connector. Check what your CRM's connector can see before you rely on it.
set up by: You, once a PropertyMe admin switches it on for youLets Claude read the property's folder, its log and the condition reports in SharePoint or Google Drive, and the email thread in Outlook or Gmail. Microsoft 365 needs a work account, and the first time anyone at your agency connects it, your Microsoft 365 administrator approves it once. Google usually needs nothing extra. No folders set up yet? You don't need any before your first job, except a home for the property's log: it works from what it can read and what you paste in, and marks every gap [MISSING]. The log is the one thing you make yourself, once (see Log it, in what it does).
set up by: You, with your own work loginIf your agency's lists live in Airtable, Claude can connect to that too, for reports. Unlike PropertyMe's, Airtable's connector can change records, not just read them. The kit only ever reads it, but that's an instruction, not a lock: so read every request, and click Allow once.
set up by: You, with your own Airtable loginOne ZIP from this guide's page. The download asks for your name and email, and we'll email you more guides. It's a plugin, an add-on file you upload into Claude once so it knows the job. Leave it zipped. Inside: the lead agent and its playbook (the kinds of difficult email and what each needs, the side by side method, what a report can and can't pull, the log layout, the checklists, the sign-off rules template and the starter prompt, which is your first message below); seven specialists (helper copies of Claude, each with one job); and three short guides. They're plain text files (written in Markdown), with no software to install and nothing to run: your own Claude reads them. Your desk checklist is a different thing: the checklist PDF on this guide's page.
set up by: YouAgency name, your name and title, phone and email, so everything it writes is yours, not mine. And who at your agency approves which emails before they go: ask Claire's Paperwork Agent for its sign-off rules template and fill it in once. You keep both in a Claude Project (the Make a Project and Add your details steps in install it).
set up by: You, with whoever sets the rules at your agencyinstall it
Do this once, ticking each step off. On a Team or Enterprise plan, your owner may need to switch some of this on first, and the menus can look different; the message for them is after the steps.
- step 1Check code execution
In Claude, select your name, then Settings, then Capabilities. Check that Code execution and file creation is on. It's the setting that lets Claude run small tools and make files inside a chat, and skills (the written procedures Claude follows, which this kit's lead is) switch on through it. It's on by default on Pro and Max, so its switch should already show on.
- step 2Switch on web search
In a chat, select + at the bottom left of the message box, then Web search, so Claude can read your state's law on the official legislation website. A tick shows beside it. No Web search there? Your Claude searches the web on its own, so skip this.
- step 3Switch off model training
Do this before any client work. On Pro or Max, select your name, then Settings, then Privacy, and switch off Help Improve our AI models: while it's on, Anthropic, the company that makes Claude, can use your chats to train its models. Its switch shows off. From now on, never rate a chat with client details in it, thumbs up or down: Anthropic says that sends it the whole chat.
- step 4Switch off memory
Do this before you use it on tenancy files. Select your name, then Settings, then Memory, and switch off Generate memory from chats and Search and reference chats, so one tenant's details stay out of another tenant's email. Both are on by default on Pro and Max. If it asks, choose Pause memory; choose Reset memory only if earlier chats held client details, because it deletes every memory for good. Both switches show off.
- step 5Check your privacy policy
Before PropertyMe or real tenancy details go into Claude, check your privacy policy. It should tell tenants and owners you use AI tools, and where their details may go.
- step 6Ask for AiMe Connect
Send your PropertyMe admin the message below, so they switch on AiMe Connect for you. It comes with PropertyMe's Advanced plan and above; PropertyMe says customers on other plans can upgrade to Advanced to get access. Not on Advanced? You can still use the kit: skip the PropertyMe steps that follow, and paste the records into your first message instead (the property's details and the email for a reply, the two reports for a side by side, your list for a report). On another CRM, ask whoever runs it for its Claude connector.
For your PropertyMe admin:Hi, can you turn on AiMe Connect in PropertyMe for me? I'm connecting it to Claude.
- step 7Connect PropertyMe
In Claude, open Customize in the left sidebar, then Connectors, and connect AiMe Connect with your own PropertyMe login. It's working when, in a new chat, you ask Claude to list what it can read in PropertyMe and it answers from your own account. Check what your CRM's connector can see before you rely on it.
- step 8Only if AiMe Connect isn't listed: ask for its link
Send PropertyMe the message below. They reply with AiMe Connect's web address, the link for Claude.
For PropertyMe:Hi, can you send me AiMe Connect's web address, the link for Claude? I'm adding it to Claude as a custom connector.
- step 9Add a custom connector
In Claude, open Customize in the left sidebar, then Connectors, and select + Add, then Add custom connector. A box opens asking for a name and a link.
- step 10Enter the name and link
Enter AiMe Connect as the name, paste the link exactly as PropertyMe sent it, and select Continue. Claude shows the sign-in settings it found, under Authentication and OAuth client.
- step 11Keep the sign-in settings
Leave Authentication and OAuth client as they are, unless PropertyMe told you to use something different; if anything doesn't match what PropertyMe sent you, stop and ask PropertyMe. Select Continue, then Add.
- step 12Sign in to PropertyMe
Sign in with your own PropertyMe login when it asks. It worked when AiMe Connect shows in your list under Connectors.
- step 13Connect Microsoft 365 or Google
In Claude, open Customize in the left sidebar, then Connectors, and connect Microsoft 365 or Google (Gmail, Google Calendar and Google Drive), whichever your agency uses, with your own work login. Microsoft 365 needs a work account, and the first time anyone at your agency connects it, your Microsoft 365 administrator (often your IT provider) approves it once: send them the message below. Google usually needs nothing extra. It's working when, in a new chat, you ask Claude to find a property's folder you know is in SharePoint or Google Drive, and it finds it.
For your Microsoft 365 administrator:Hi, we're connecting Claude to our Microsoft 365. Before any of us can connect, a Global Administrator needs to approve it once for the whole agency. The quickest way: sign in to Claude, open Customize, then Connectors, connect Microsoft 365, and tick the box to grant access for the whole organisation. No Claude account? Anthropic's help article on setting up the Microsoft 365 connector shows how to do it in Microsoft Entra instead. Thanks.
- step 14Only if Google says Access blocked
Your Google Workspace admin limits outside apps. Send them the message below. Once they've set Claude as a trusted app, connect Google again.
For your Google Workspace admin:Hi, Claude's Google connectors show Access blocked for our accounts. Can you please set Claude as a trusted app in our Google Admin console, under Security, then Access and data control, then API controls, then Manage third-party app access? Thanks.
- step 15Only if your lists live in Airtable
In Customize, then Connectors, connect Airtable with your own Airtable login. Unlike PropertyMe's, Airtable's connector can change records, not just read them. The kit only ever reads it, so read every request, and click Allow once.
- step 16Only on Safari on a Mac: keep ZIPs zipped
Open Safari, then Settings, then General, and untick Open "safe" files after downloading. The box shows empty, so the kit stays a ZIP when you download it.
- step 17Download the kit
On this guide's page, in the "in this folder" list, select ZIP beside the agent kit. The download asks for your name and email, and we'll email you more guides. claire-paperwork-agent.zip is now in your Downloads folder. Leave it zipped.
- step 18Open Plugins
The kit is a plugin, an add-on file you upload into Claude once so it knows the job. In Claude, open Customize in the left sidebar, then Plugins. Not there? Menus move: search Claude's help for plugins.
- step 19Upload it
Select Add, then Upload plugin, and choose the ZIP file itself, not a folder. Only a folder after downloading? Right-click the folder, choose Compress, and upload the ZIP that makes.
- step 20Check it's on
Check that Claire's Paperwork Agent shows in your plugins list, switched on. If it's off, switch it on.
- step 21Make a Project
A Project is a space in Claude that keeps the same instructions for every chat in it, so you give your details once. Select Projects on the left, then + New Project, and name it, for example with your agency's name. Work under more than one tenancy law? Make one Project for each state.
- step 22Add your details
In your Project, select Set project instructions. Paste your agency's details (agency name, your name and title, phone and email) and your sign-off rules, then select Save instructions. Claude now uses them in every chat in the Project. Nothing more to switch off here: with memory off in Settings, then Memory (the Switch off memory step), Anthropic says Claude won't use memory or make new memories.
- step 23Only if your sign-off rules aren't written yet
Start a chat in your Project and ask Claire's Paperwork Agent for its sign-off rules template. Fill it in with whoever sets the rules at your agency, then paste it into the Project's instructions.
- step 24Test it
Start a new chat inside your Project and paste your first message from below, with this first line: Show me the inspections booked for next week, across my properties. It's working when Claude asks for anything missing in one message, asks Allow once before it reads, then stops with a table that says when it was read and names the record behind each row. No CRM connector yet? Paste a short invented history and an invented upset email instead, and ask for a reply: it's working when Claude asks for anything missing in one message and hands back a draft with [MISSING] marks and its check.
That's all you need, on claude.ai or the Claude desktop app, with no coding: in chat, Claude does each of the seven specialists' jobs itself, one after another (they may show greyed out in chat; that's expected). In Claude Code, Claude's app for developers, the kit is built to run them as their own agents; you don't need it. Menus move: if yours look different, search Claude's help for plugins, projects, connectors, code execution or web search. On a Team or Enterprise plan, if a setting or connector above is missing, send your Claude admin this message: "Hi, I'd like to add Claire's Paperwork Agent, a free plugin for the work around our tenancy paperwork (difficult emails, the property log, condition report comparisons and reports), to my Claude. It comes as a ZIP file. Can you please allow plugins, switch on Code execution and file creation and web search, and add our CRM's connector (AiMe Connect for PropertyMe; if it isn't listed, add it as a custom connector with the address PropertyMe gives us), Microsoft 365 or Google, whichever we use, and Airtable if we use it, so I can connect my own account? Thanks."
your first message
Copy it and send it in a new chat inside your Project, on the Claude website or the desktop app. One job, one chat. Swap the first line (highlighted) for each job and keep every other line. For a reply: "Help me reply to the tenant's email below, for [address], [suburb] [state].", then paste the email under the last line. For a side by side: "Compare the entry condition report with the exit report, side by side, for [address], [suburb] [state]." For a report: "Show me the leases coming up for renewal in the next 60 days, across my properties." To log it: "Log this: [one line on what you did], for [address], [suburb] [state]." Giving it figures? Say where each came from. Not on PropertyMe? Write your CRM's name instead.
Use Claire's Paperwork Agent. Check PropertyMe, the property's folder, its log and my inbox first, as the job needs. Mark anything you can't confirm as [MISSING], and name the Act, the section or clause, and the date for any legal point. Drafts only. I'll check it and send it myself, and I'll say OK before anything goes in the log.
the four rules it runs on
- It reads first. It's built to read the history first: the property in PropertyMe, its folder and its log, and the email thread in your inbox. Anything it can't confirm is marked [MISSING]: find that fact, never ask it to make one up.
- Claude drafts, you decide. Keep it to drafts only, and send it yourself, from your own email. When it asks permission, read what it's about to do, then click Allow once, never Always allow. If it ever asks to send an email, don't let it. What's fair wear and tear, and what goes in a claim, is your call.
- Every finished job goes in the log. You make the log once, in the property's top folder. Claude writes each entry with your name, plus one short line for your CRM note, for you to paste. Check your name and the facts, then say OK. Log entries are never edited or deleted: a mistake gets a correction entry. Before you close the chat on a finished job, make sure it's in the log.
- Check what your connector can see. Ask for the leases coming up for renewal, the inspections booked, or the open jobs. For any other report, ask it, and check what your connector can see before you rely on it. Arrears, rent, ledgers and bonds come from you, with their source and date.
keep it safe
- A person decides and a person sends. Claude drafts, and the kit tells it never to send anything. That's an instruction, not a lock: your email connector may be able to send, so if it ever asks to send an email, don't let it. The right person at your agency approves under your own sign-off rules, and you send it yourself.
- Check your privacy policy first. Before PropertyMe or real tenancy details go into Claude, check your privacy policy. It should tell tenants and owners you use AI tools, and where their details may go.
- Work accounts only. What you paste and what Claude reads goes into your own Claude account, never to us. Use Claude on your work email, never a personal account.
- Where tenants' details go. Anthropic, the company that makes Claude, says in its privacy policy (anthropic.com/legal/privacy, effective 10 September 2026, read 17 October 2026) that personal data you put in a chat is collected, and that personal data is transferred to its servers in the US, or to other countries; its trust centre lists the other companies it goes to and where they are. Its privacy centre (How long do you store my data?, read 17 October 2026) says that on Pro and Max, a chat you delete leaves your history straight away and its storage within 30 days, though it may keep chats longer: where the law requires or to resolve disputes; up to 2 years if its safety systems flag a chat; up to 5 years, de-identified, if Help Improve our AI models was on; and 5 years for a chat you rated with thumbs up or down. So a tenant's details in a Claude chat may go overseas: check your own privacy policy says so before you put real records into Claude, by a connector or by pasting.
- Switch off model training first. It's the Switch off model training step in install it, before any client work: on Pro or Max, select your name, then Settings, then Privacy, and switch off Help Improve our AI models. Anthropic says Team and Enterprise chats aren't used for training by default.
- Switch off memory first. It's the Switch off memory step in install it: select your name, then Settings, then Memory, and switch off Generate memory from chats and Search and reference chats before you use it on tenancy files, so one tenant's details stay out of another tenant's email. On Team or Enterprise, your Claude admin decides whether memory is available.
- Never rate a client chat. No thumbs up or down, on any plan. Anthropic, the company that makes Claude, says that sends it the whole chat.
- Allow once, never Always allow. Read what it's about to do each time. Always allow stops it asking you. If it's not what you asked for, don't allow it.
- Read only, everywhere but the log. The kit never writes in PropertyMe or Airtable, and never saves, moves, renames or deletes a file. The only thing it adds is one new log entry, after you say OK.
- Nothing secret in a chat. No passwords, no card numbers, no ID photos. The kit never needs one.
- One job, one chat, with memory off. Another property means another chat, and with memory and chat search off, one tenant's details stay out of another tenant's email.
- Your figures, with their source. Arrears, rent owing, ledgers and bonds come from you, from your trust ledger, with the date. Claude never pulls or works them out.
- Your state's rules come first. Tenancy law differs by state and changes, and the kit checks your own state's law. Any law shown in our video is ACT law; rules differ in other states. When in doubt, ask your licensee or compliance person. This kit is not legal advice.
the full system
The kit is the playbook. Around it sits your own setup: your Claude, your CRM, your documents and email, your lists, and your agency's own rules. Your notices and letters still come from your CRM's templates. It all runs on the Claude website or the desktop app, with no coding. Here's all of it, in the order to set it up.
Reads the records, drafts the difficult email, sets the reports side by side, pulls the report and writes the log entry
Claude on a paid plan (Pro or Max; Team and Enterprise too), in chat on claude.ai or the desktop app · paid: yes, your Claude planThe steps and rules Claude follows on every job
This guide's free kit: a Claude plugin with a lead agent and seven specialists, written instructions with no code · paid: noThe property, its contacts, jobs, tasks, inspections and lease renewals; your notice and letter templates stay here
PropertyMe's AiMe Connect (read only, PropertyMe says), or your own CRM's Claude connector · paid: comes with PropertyMe's Advanced plan and aboveThe property's folder, its log and its condition reports
SharePoint or Google Drive, through Microsoft 365 or Google · paid: your existing planThe thread the difficult email answers, and the reply you send yourself
Outlook or Gmail, through the same connector · paid: your existing planAny lists your agency keeps outside the CRM, for reports
Airtable's own connector, read only by the kit's rule · paid: your existing planYour state's tenancy law, read at the source
Web search, on your state's official legislation website · paid: noWho signs off which emails, your rules for emails, your log layout and file names
The kit's sign-off rules template, kept in a Claude Project (a space in Claude that keeps the same instructions for every chat in it) · paid: noWhat your tenants and owners are told
Your agency's privacy policy, saying you use AI tools and where their details may go · paid: nosetting it up, in order
- step 1Claude
A paid plan on your work email, with code execution (the setting that lets Claude run small tools and make files inside a chat; skills switch on through it) and web search on, and Help Improve our AI models, memory and chat search off.
- step 2Privacy
Check your privacy policy tells your tenants and owners you use AI tools, and where their details may go, before PropertyMe or real tenancy details go into Claude.
- step 3Your CRM
A PropertyMe admin switches on AiMe Connect for you; then connect it in Claude, under Customize, then Connectors. Ask Claude to list what it can read, and check what your CRM's connector can see before you rely on it.
- step 4Documents and email
Connect Microsoft 365 or Google with your own work login. On Microsoft 365, your administrator approves it once.
- step 5Your lists
Only if your agency's lists live in Airtable: connect Airtable with your own login. Its connector can change records, so read every request and click Allow once; the kit only ever reads it.
- step 6The folders and the log
No folders to build before your first job, except a home for each property's log: the kit works from what it can read and what you paste in. The log you make yourself, once per property, in its top folder: a spreadsheet or list with five columns (date, who, what was done and why, who approved it, where it's filed) if you want Claude to add rows on your OK, or a document you paste entries into. The kit never creates a file. As you go, one folder per property (suburb, then the address, then the right subfolder) keeps things easy to find.
- step 7The kit
The kit is a plugin, an add-on file you upload into Claude once so it knows the job. Upload the ZIP once, under Customize, then Plugins, and check it's switched on.
- step 8Your rules
Make a Claude Project: select Projects on the left, then + New Project, then Set project instructions. Paste your agency's details and your filled-in sign-off rules (ask Claire's Paperwork Agent for its sign-off rules template), then select Save instructions, and start each job's chat inside it. Work under more than one tenancy law? One Project for each, and never mix the two.
- step 9A test job
Ask for the inspections booked for next week, and check the table says when it was read and names the record behind each row. No CRM connector yet? Paste a short invented history and an invented upset email instead, and ask for a reply. Then try a reply to an email you've already answered, and compare it with what you sent, without sending anything.
things to get right
- Privacy. Your privacy policy tells tenants and owners you use AI tools and where their details may go, client work stays on your work account, model training and memory are off, and no chat with client details in it is ever rated.
- A person signs off. Claude reads, drafts, compares and checks. A person at your agency checks the facts, makes the calls, approves and sends.
- Read only. Reading and drafting only. Allow each action once, and never Always allow. The only thing the kit adds anywhere is one log entry, after you say OK.
- The record. The log is only ever added to. A mistake gets a correction entry, never an edit.
- The law. Every legal point names the Act, the section, clause or regulation, and the date, read on your state's official legislation website. This is not legal advice.