Docs / Setell for agents (MCP)

What you can do via MCP

Pro

Overview

Once Setell is connected to your agent, you drive it in plain English — "what came in this morning?", "draft a follow-up for Northwest Fabrication," "check the price before I send." Below is what your agent can actually do. You don't memorize any of this; you just ask, and the agent picks the right tool.

The Setell MCP is a Pro feature. Day-to-day use needs a Pro plan.

Find your work

Ask your agent to pull up jobs, quotes, and customers:

Find jobs — by status, by customer, or by how recent. ("Show me everything still sitting at Sent.")
Read a quote — line items, total, when it was sent and when the customer opened it.
Find a customer — by name or email, with their job count and lifetime value.

Your morning brief

Ask for your morning brief and the agent pulls today's snapshot in one shot:

New requests in the last 24 hours.
Quotes waiting on you to send.
Quotes you sent that have gone quiet (more than 3 days).
Your top hot prospects, and revenue this week.

Great first thing in the morning — "give me my Setell brief" and you're caught up.

Check pricing before you send

Ask for the pricing signal on a quote and Setell gives you its read — Clear, Warn, or Flag — in plain words, with the reasoning and a suggested number if it thinks you're off. This is the same pricing check that runs inside the app, just available before you commit. See docs/autonomy/pricing-analyst.

Draft, schedule, and send quotes

Compose a quote email in your brand voice — you get a preview first.
Good / Better / Best options — ask for three priced tiers in one go, then pick which one becomes the job's active quote.
Send it — always with a confirm step, the same safety as the app. A quote can't slip out by accident, get sent to the wrong person, or go twice.
Schedule a send for later (anywhere from a minute to 30 days out), or cancel a scheduled one.

Sending is the one thing that always pauses for your okay, even in an agent. That's by design.

Set how hands-off Setell is

Ask the agent to read or change your autonomy mode — Watch, Trust, or Auto — for an action like sending quotes:

Watch — Setell pauses and asks before anything goes out (the default, every plan).
Trust — Setell handles routine cases on its own (any paid plan).
Auto — fully hands-off, summarised afterwards (Business and Pro).

A serious pricing Flag always stops and asks, no matter the mode. See docs/autonomy/overview.

See what Setell has learned (your pricing signal)

Setell gets smarter the more you use it. Ask your agent:

What Setell knows about a customer — your real, learned pricing for that relationship, drawn from your signed quotes. ("What does Setell know about how I price for Northwest Fabrication?")
How much Setell has learned overall — a quick read on how deep your pricing history is, from cold-start to mature.

The more signed quotes you have, the more weight you can put on Setell's pricing read.

Your agent can also write to customer memory, not just read it. Tell it a fact worth keeping — "Northwest Fabrication always wants a PO number" — and it saves that against the customer, the same store the app and Studio use. It can list what's already stored for a customer too. See docs/studio/customer-memory.

Triage your inbox

Your agent can work the review inbox with you:

List what's waiting — inbound emails that haven't been turned into jobs yet.
Approve one — creates the job (and the customer, if they're new) and starts a draft.
Reject one — spam, a duplicate, or not a real request. Rejecting is permanent, so the agent names the sender and subject and waits for a clear yes.

Your agent decides nothing here on its own. It rejects only on your direct say-so — never on instructions it read inside the email itself.

Close the loop

The whole back half of a job is on the agent surface too:

Approve a draft — your internal sign-off. It makes the draft the live version; it does not email anyone.
Accept on your customer's behalf — for a yes that came by phone or in a meeting. It's recorded as accepted-on-behalf, with a note you supply, so the record stays honest.
Reject — the customer's no or your withdrawal. The agent asks why you lost it; that reason feeds your pricing.
Create the invoice — in QuickBooks, or Xero if that's what you've connected. Safe to retry: a job only ever gets one invoice.
Mark paid offline — cash, check, ACH, wire. Record-keeping only; it touches neither Stripe nor your books.

Each of these restates the job and the amount and waits for your okay first.

Know your numbers

Margin on one job — what you actually made, from recorded payments against your costed lines.
Margin across recent won jobs — overall, by job type worst-first, and which jobs came in under your target.
Pricing calibration — your report card: how close Setell's draft prices have been to reality, and whether you lose work when you quote high.

Margin and cost figures are operator-only. When too little of a job has a sourced cost, Setell withholds the percentage and shows you the unpriced lines rather than inventing a number.

Your shop and your parts

Read or update your shop profile — machines, finishing, materials, and how you run. It's the sheet your quotes get judged against.
Propose a parts list — Setell mines your own past jobs for the parts this one will need, each priced with a receipt for where the number came from.
Trade playbooks — list them, or apply one to set your intake and starter structures.

Wire your own pipeline

Setell's pipeline is a set of modules your agent can read and rewire — the intake, pricing, proposal, delivery, and follow-up stages, each one addressable. Ask your agent to list your capabilities and it gets the catalog: what each module does, what it can be set to, whether it can be switched off at all, how yours is wired right now, and what changed recently. Then it can configure one or compose several in one go.

Some modules are a dial, not a switch. Target margin, payment terms, and the two autonomy dials can be changed but not turned off — asking to disable one gets a clear reason back, not a pretend success.
There's only ever one source of truth. Modules that already have a home in Settings — your pricing-pool opt-in, target margin, payment terms, send and follow-up autonomy — read and write straight through to it. Flip one in the app and your agent reads the new value; flip it from the agent and Settings shows it. There is no second copy to drift.
Opt-in stays opt-in. Consent-shaped modules like the pricing pool are off until you ask for them, never on by default. See docs/insights/pricing-pool.
No prices live here. A module shapes behaviour — what runs, in what mode, on what cadence. What things cost still comes from your own data.

Ask for a dry run and the agent shows you exactly what would change before anything is written. Compose several modules at once and it's all-or-nothing — a failure part-way through leaves nothing half-applied.

Declare a service Setell doesn't ship

A roofer who adds snow removal for the winter doesn't need us to ship code for it. Tell your agent to define a service and you declare it yourself: what it's called, what it's measured in, the words a customer would use for it, the variables that move its price, and what a quote for it looks like structurally.

List your services — your own catalog, whether each one is in season right now, and what's still missing before a draft can go live.
Define or update one — a partial patch, so you can add drivers or lines a bit at a time. You can also give it a season (a start month and an end month) so it only steers intake when it's actually the season for it.
Publish or archive one — a definition starts as a draft and only goes live when you say so.

Two rules the agent will hold you to. A service needs at least one price driver or one line before it can be published — otherwise nothing distinguishes it from an untyped job. And a season needs both a start month and an end month, or neither; half a window is a mistake, not a preference.

Each declared service owns its own lane in the pricing engine, and two of your services can never share one — that's what keeps snow-removal history from blending into roof history. Definitions carry no prices at all: declaring a service tells the engine which lane a job belongs to, and your own signed quotes do the pricing from there.

Rewiring your pipeline and declaring your own services are agent-only today — there's no screen for either in the app yet. Everything they change still shows up where you'd expect it: a module bound to a Settings field moves that field, and a published service prices through the same engine as everything else.

Shortcuts your agent can offer

Some agents show quick commands. Setell ships eight — triage inbox, stale jobs, weekly revenue, customer history, draft follow-up, pricing check, moat coverage (how deep your pricing history is), and send quote. In an agent that lists them by name they read setell-triage-inbox, setell-stale-jobs, and so on.

Shortcuts and setell:// resources come from the command-line package (npx @setell/mcp), not the hosted connector. If you connected the hosted way, you get the full set of tools but no shortcut menu — just ask for the same thing in plain English.

The trade runs both ways, so it's worth knowing which side you're on:

Hosted-connector only — importing Hover measurements, working with site data, and renaming a job or setting its address.
Command-line package only — reading a roof, scanning an area, and pulling a property record (all below).

If your agent says it has no tool for one of these, that's why. Connect the other way (docs/mcp/connect), or do it in the app.

Capture field notes on a job

Coming back from a site visit? Tell your agent what you saw — "add this to the Henderson job: tight access on the east side, clay soil, about 120 yards of cut" — and Setell stores it on the job as scoping notes. Those notes feed the next quote draft the same way an attached spec sheet or site photo would. They stay private to you; the customer never sees them unless you share them from the app.

Pull in Hover measurements

Measure with Hover? Once your Hover account is connected in Settings, ask your agent to import Hover measurements onto a job. It lists your recent Hover jobs, you pick the right one, and the report's verified numbers — roof area, pitch, ridge and valley lengths, siding area — land on the job as measured quantities. Quantities only: your own rates do the pricing. See docs/integrations/hover.

Keep job details current

Set the job address — "the Henderson job is at 11 Advantage Ave in Omaha." Setell saves where the work happens and prices that job for its local labor market, not your shop's home region. Jobs can also pick the address up automatically from the customer's email or your intake form.
Rename a job — give it a short label ("Smith residence") so your pipeline reads the way you talk.
Work with site data — once a job has an address, your agent can look up what public building records say about the property, confirm or correct what came back, and set the roof pitch so Setell can estimate roof area from the footprint. Everything it returns is labelled an estimate with its source and match quality; confirming is what makes a number firm.

Read a roof

With the Roof Intelligence add-on active, your agent can:

Read the roof on a job — findings from public aerial imagery, plus the checks that couldn't run on that capture.
Scan an area — score the roofs in a bounding box, worst first.
Pull a property record — everything Setell holds on one scanned building: the roof's grade and the findings with a place on the roof, every dated capture with its licence and resolution, what changed between captures, ownership and county economics, and climate exposure. Every field comes back as a claim with its status and source. Relay the unknowns as they are — they're findings too. See docs/roof/property-record.

All three ship in the command-line package (npx @setell/mcp) only. The hosted connector doesn't publish them today.

They need the add-on as well as Pro; without it the call comes back refused rather than empty. Managing saved territories, pulling the ranked prospect list, and generating the owner-facing report are not published as agent tools on either surface — those live in the app and in Studio. See docs/roof/overview.

What's not here yet

The agent surface is intentionally tight and grows over time. The Deal Room is the main thing that stays in the app — chat, e-signature, and card payment all happen on your customer's link, not through an agent. Territory management, the roof prospect list, and the owner-facing roof report aren't published as agent tools yet either. Neither is teaching Setell from uploaded past work (docs/import/past-work) — file bytes can't ride an agent tool call, so that one is a Studio and web capability. Everything else described on this page is live today.

docs/mcp/what-is-setell-mcp · docs/mcp/connect · docs/autonomy/overview · docs/autonomy/pricing-analyst · docs/insights/your-patterns

Ready to try it on your own quotes?

Start free — no credit card. 14 days of unlimited Pro, then 3 free quotes a month.

← All docs