Full analysis
Let relea carry the screen.
Hand it a listing, a PDF, or a pile of addresses. It builds the model, scores your buy box, and comes back with the deals worth your time. That is how you scale underwriting without adding hours.
Claude · ChatGPT · Cursor
Your AI assistant for
real estate investing.
Your AI can reason through a deal. On the open web it rarely reaches the county record, the loans recorded against the property, or the unit rents in the MLS. Connect relea and it can, with a source on every number. Let relea run the full analysis, or keep your own model and ask relea for the facts.
Included on every plan, Free too. Three listings free, then from $3 a listing. Reading a deal that's already in your account is free.
Inside your process
Or ask it for the facts.
Keep the spreadsheet, the memo, or the model your team already trusts. Ask relea for the county record, the tax history, the recorded loans, the rents, and the documents. Your process stays yours. The inputs get sounder.
AI only vs AI + relea
Same AI. Better evidence.
We gave the same AI model the same underwriting request on 10 live multifamily listings, once with web search alone and once with relea connected. Here is what the connected run could see.
Recorded loans
9 of 10 deals, against 0.
Connected to relea, the agent pulled the loans recorded against 9 of the 10 properties. On the open web alone, it found none. A recorded loan tells you where the seller's floor likely is.
Tax history
About twice as many years.
11 years of tax history on average, against 5. Enough to see the trend and the bill a buyer will actually inherit.
MLS unit rents
The rent roll behind the listing.
relea surfaces the unit-level rents and seller expense lines in the MLS record, which often don't show on the public listing page.
Fewer dead ends
Half the web calls.
About 16 web calls per deal instead of 29, and half as many that failed or came back empty. The agent spends its effort on research relea doesn't cover.
The rent roll your AI couldn't see.
A 1961 six-unit in Pico-Union, Los Angeles, listed at $1.6M. Same model, same prompt, same investor terms: 25% down, 7%, and hurdles of a 6.5% cap rate and 1.25 DSCR.
| AI only | AI + relea | |
|---|---|---|
| Rents | 2 of 6 found on rental sites; the other 4 guessed at $1,300 | All 6 from the MLS unit table: $11,906 a month |
| Seller's expenses | None found | Trash, water, house electric, and insurance from the listing |
| Seller's debt | Not found | A $1.1M loan recorded in 2021 |
| Year-one NOI at ask | $49,632 | $73,261 |
| Most to pay at your hurdles | $790,000 | $1,062,000 |
| Recommended offer | $750,000 | $1,025,000 |
Both runs said pass at the asking price. One said there was no deal to discuss. The other showed where the line is: your number is about $1.06M, and the recorded loan suggests the seller needs somewhere near $1.08M to $1.10M. The guessed rents were about 35% low, and the most-to-pay came out $272,000 lower.
The connected run still did its own work. It overrode relea's market-rent estimate with live comps, reset the property tax under Prop 13, and checked the city's rent rules. relea brings the record. Your AI brings the judgment.
Live listings run on October 27, 2026. Both runs used the same model, prompt, and budget, and neither saw the other's work. A grader who didn't know which answer used relea preferred the relea answer on this listing. The figures above describe one deal, not a typical result, and nothing here is investment advice. Every relea field now says whether it is a record or an estimate, so your assistant can tell which is which.
1
Connect
Add a custom connector.
Sign in. No API key.
Claude and ChatGPT both take the same URL. Paste it, then sign in to relea when the window opens. The menus below are what you will click.
https://mcp.relea.ai/mcp
Claude
Settings, then Connectors, then Add custom connector.
-
1
Open Connectors. Choose Add custom connector.
-
2
Name it relea. Paste the URL.
-
3
Connect, then sign in. Claude is ready.
ChatGPT
Settings, then Connectors. Developer mode is the switch that unlocks a custom connector.
-
1
Open Connectors, then Advanced. Turn on developer mode.
-
2
Add the connector. Paste the same URL and create it.
-
3
Sign in to relea. Start a chat and ask.
Cursor and other clients that offer a custom connector use this same URL and the same sign-in. An icon on the ChatGPT form is optional.
2
Ask
Then ask, in the chat you already have open.
County and tax, with sources
The assessor record before you open a spreadsheet.
Get the county assessor classification and the tax history for 123 Main St, Chicago. Cite the source for each figure.
The rent roll, with its basis
What the units bring in today, and which numbers are records.
What rent does each unit in this listing bring in today, and where does each figure come from? Separate records from estimates.
Recorded loans and liens
The evidence behind a negotiation.
What loans are recorded against this property, when, and for how much? What does that suggest about the seller's floor? Use relea.
Why this bid
releaOffer names every factor and its dollar impact.
Why is relea's bid below the ask on this listing? Show me each factor, its dollar impact, and the evidence behind it.
Your walk-away, on your terms
A what-if that saves nothing to the deal.
What's my walk-away at 30% down, 6.75%, and a 1.20 DSCR, and how far is it from relea's bid and the seller's loan?
The full model, or one check
Let relea underwrite it, or check one claim against your own numbers.
Pull the OM and county tax record for this property and check the seller's NOI claim against relea's underwriting.
Ask your assistant to cite relea's source for each figure. Every field says whether it is a record, a figure derived from records, an estimate, or a default, so you know which numbers to check.
What it costs
A few dollars a deal. Every question after that is free.
In the Los Angeles test above, one listing's worth of records moved the most the investor should pay by $272,000. Here is what that listing costs in relea.
One credit per new property
Submitting an address relea hasn't seen starts the analysis and uses one credit the first time data comes back. A listing that fails before returning anything costs nothing.
Reads are free
Once a listing is in your account, your assistant can ask about it as often as it likes. Reading it never starts a new analysis or uses a credit.
No subscription needed
Three listings free, no card. Then $5 for one, $20 for five, or $74.99 for 25. Credits never expire. Plus and Pro bring a listing down to $1.96 and $0.99.
Connecting Claude or ChatGPT is included on every plan, Free included. A scheduled job can submit new addresses without confirming each one. See all plans.
Scale the screen
Point it at a list. Hear about the ones that fit.
Spreadsheet
Forty addresses, one ask.
Here is my Excel from a wholesaler. Run them through relea and tell me which clear my buy box: 7%+ cap rate, DSCR above 1.25, under $150k per unit.
CRM or database
Write the answers back.
Every Monday, pull new deals, underwrite them in relea, and write cap rate, NOI, releaOffer, and the top risks back to each row.
A watch list
Only ping me on a fit.
Take my Redfin favorites, analyze each one, and message me when any meets my criteria or drops in price.
Your own model
Just the sourced inputs.
Do not rebuild my underwriting. Return property tax, open liens, year built, and the rent comps for these addresses, and cite the source on each.
The first fields come back while the rest of the analysis is still running, so a batch is useful before it finishes. Ask for only the fields you need, and every number has a source your assistant can cite.
What you can ask for
Everything relea can return, on request.
Ask by name or by category. Every field comes back with its source and a basis, so your assistant can tell a record from an estimate. This is the short map. The full catalog, with every field and its units, is one page over.
| Category | Examples |
|---|---|
| identity, listing, listing_history | address, price, days on market, price cuts, relists |
| county_assessor, tax | class and land-use codes, assessment history, tax history, underwritten property tax |
| ownership_and_sales | owner, sales and transfer history, ownership length, owner-occupied |
| seller_equity_and_liens | open liens, recorded loans, equity estimate, foreclosure flags |
| building_systems, characteristics | roof, HVAC, plumbing, electrical, square feet, year built, zoning |
| units_and_rents, comps | unit mix, in-place rents from the listing, releaRent estimate per unit type |
| income, expenses, assumptions | in-place and stabilized lines, vacancy, management fee, taxes, insurance |
| deal_metrics, returns | NOI, cap rate, GRM, DSCR, cash flow, cash-on-cash, IRR, equity multiple |
| offer | releaOffer bid and range, the named motivation, market and condition factors with dollar impact, capex ask, seller equity, and your walk-away |
| risks_and_diligence, condition_and_risk, capex | four-tier risks, diligence checklist, renovation plan |
| documents | OM, rent roll, T-12, county pages, tax bills, recorded documents, photos |
Connected assistants can also read the catalog as relea://catalog/fields.
No sign-in connector?
If your client cannot do that OAuth sign-in.
Some tools only accept a token in a config file. Create one in relea under Settings, Integrations, then send it as a bearer token. Disconnect an app from the same page. Claude and ChatGPT do not need this.
{
"mcpServers": {
"relea": {
"type": "http",
"url": "https://mcp.relea.ai/mcp",
"headers": { "Authorization": "Bearer relea_pat_REPLACE_ME" }
}
}
}
Claude Code can sign in without a token: claude mcp add --transport http relea https://mcp.relea.ai/mcp. Add --header "Authorization: Bearer $RELEA_TOKEN" only when you are running it from a script.
ChatGPT said it could not discover settings
That usually means the first lookup raced a restart. Open the connector again. When the advanced OAuth fields are filled in, create it and sign in. The sign-in screen is relea.