Real Estate Stack · Chapter 02 of 10

CRM from zero: choose it, set it up, connect it to Claude

A CRM you never configured is just an expensive contact list. This chapter picks a product by tier, builds the pipeline, tags, and consent fields from nothing, and sets the rule for what Claude may touch and what a human must click.

Why this chapter

Most agents buy a CRM, import a spreadsheet, and stop. The CRM never gets pipeline stages, tags, or a consent field, so it can't tell you who to call today and it can't tell a texting tool who is legally textable. This chapter builds that structure first, then adds Claude on top of it: drafting, summarizing, and proposing, never sending or saving without a human click.

Choosing by tier

Nine CRMs a solo agent or small team is likely to shortlist in 2026, priced and checked for AI features, API access, an official MCP server (a vendor-run connector Claude can talk to directly), and texting-compliance tooling. Prices dated 2026-09-03.

CRM	Price (2026-09-03)	Native AI	API / Zapier	Official MCP	Compliance features	Evidence
Follow Up Boss	Grow $69/user/mo ($58 annual); Pro $499/mo for 10 users; Platform $1,000/mo for 30	Smart summaries, smart messages, predictive lead prioritization	Open API + webhooks; Zapier yes	No. Third-party servers only (Ace AI, a free community server)	Not found (no FUB-specific consent field or 10DLC page)	verified [1]
Lofty (formerly Chime)	Not published; four tiers, all "Request Pricing"	AI Task Prioritization, Lead Scoring, Sales Agent, Homeowner Agent	~94 REST endpoints + webhooks; Zapier yes	No. Community server wraps the REST API	A2P 10DLC registration flow documented in help center	claim [4]
kvCORE / BoldTrail	Not published; third-party estimate solo ~$499/mo, teams $499-999/mo	Smart CRM lead scoring, Marketing Autopilot, HomeSearch AI	Public + Partners API, signed webhooks, gated to paid tiers	No official MCP found	TCPA opt-in checkbox locked by default; opt-out text every 5 messages or 15 days	report [7]
GoHighLevel	Starter $97/mo; Unlimited $297/mo; Agency Pro $497/mo; AI Employee add-on $50-97/mo	Conversation AI, Voice AI, Content AI, AI Employee	Native API + webhooks; Zapier yes	Yes. Official MCP, ~39 tools, free on Unlimited/SaaS Pro	A2P 10DLC via built-in LC Phone brand/campaign wizard	verified [9]
Sierra Interactive	Starter $299.95/mo annual (1 user); Essential $399.95 (3); Growth $599.95 (5), 12-month terms	Lead Engage AI nurture add-on, $199/mo unlimited leads	Official API docs; Zapier not confirmed	Not found	Automatic opt-in/opt-out language, A2P assistance, delivery reporting	verified [12]
Wise Agent	CRM $49/mo ($42 annual); CRM+WiseSocial $69/mo; WiseText $11/mo + $80 registration fee	None mentioned on the pricing page	API key exists, full credentials approval-gated; Zapier yes	No	$80 WiseText fee corresponds to A2P 10DLC registration	verified [14]
HubSpot Free CRM	Free, up to 2 users, 1,000 contacts; Starter paid tier from $7/seat/mo	Basic chatbot/live chat free; Breeze AI agents need paid Credits	Full public REST API + webhooks; Zapier yes	Yes. Official Remote MCP Server, GA April 13, 2026, OAuth, read/write	Standard unsubscribe management on paid Marketing Hub tiers	verified [15]
Twenty (open source)	Pro $9/user/mo annual; Organization $19/user/mo; self-hosted free	Native AI chat, custom AI agents, workflow automation credits on paid tiers	REST + GraphQL, webhooks on Pro/Organization	Yes. Native MCP server ships in every paid Cloud workspace	Enterprise-only: single-tenant isolation, IP allow-list, SCIM, audit logs	verified [17]
Notion as CRM	Free $0; Plus $10/user/mo; Business $20/user/mo	Notion AI chat/autofill on Free/Plus; full AI + Agent on Business/Enterprise	Public API + webhooks on paid plans; Zapier via connected properties	Runs its own workspace MCP, but it is not a CRM feature	None. No SMS/email compliance layer, no 10DLC, no consent fields	verified [18]

Read the "Official MCP" column literally. A Zapier or community listing that says "MCP for [product]" is usually Zapier's own generic gateway wrapping that product's existing Zapier actions, not a server the vendor runs and supports. verified [1][7][14] show that pattern for Follow Up Boss, kvCORE, and Wise Agent; GoHighLevel, HubSpot, and Twenty are the three in this table with a vendor-run server as of 2026-09-03.

If you are choosing today: Notion or Twenty if you want a free or cheap start and are comfortable building your own pipeline template. Follow Up Boss or Lofty if you want a real-estate-specific product built around lead speed and texting compliance. GoHighLevel if you are running a brand, not a personal book of business, and want an official Claude connector out of the box. HubSpot's free tier is the one to open first if all you need today is a place to put contacts and an official MCP server, without committing to a real-estate vertical product yet.

The relationship database vs volume engine split

One CRM rarely does both jobs well. A working Arizona buyer's agent's procedure operator [19] runs two systems side by side, for a large cold list plus a personal book of clients, and the split generalizes past that one deal:

Relationship database (Lofty or Follow Up Boss): people you actually talk to. Active buyers and sellers, past clients, sphere. Smart plans, transaction milestones, and MLS/IDX tooling work well here because the list stays small and every record gets attention.
Volume engine (GoHighLevel): a large, mostly cold list under a separate brand. GoHighLevel is built for exactly this: a standalone domain, phone number, and funnel, holding thousands of records with suppression logic, without dragging down the working database's views and smart plans.

The boundary is the hand-raise. A contact lives in the volume engine as a subscriber until they reply, book a call, or otherwise show real interest, at which point they get created in the relationship database and you work them personally. Do not mirror the whole cold list into both systems: that doubles the work and muddies who has actually consented to what. operator [19]

Reasons this split holds up beyond one operator's list: brand separation (a volume campaign under a project or team name is a different first-touch face than "Jane Smith, Realtor"), consent hygiene (provenance for an unconsented mass is easiest to track in one place), and blast-radius containment (if a cold-list campaign damages a sending domain or a phone number, the damage stays in the volume engine and never touches the identity you use with real clients). operator [19]

The setup procedure from zero

Do this in order. Skipping the fields and tags is why most agent CRMs never produce a usable daily call list.

Pipeline stages

Build two pipelines, not one. A buyer and a seller move through different work, and one shared pipeline hides which kind of appointment is overdue.

Buyer pipeline	Seller pipeline
New lead	New lead
Contacted / qualifying	Contacted / seller consult set
Buyer agreement signed	Listing agreement signed
Actively touring	Prep / photos / staging
Offer written	Live / active on MLS
Under contract	Under contract
In escrow / build period	In escrow
Closed	Closed

A working volume-engine build adds a longer pre-buyer pipeline in front of this: New (imported) → Emailed → Replied / Engaged → Confirmed still looking → Quiz completed → Call booked → Handed to agent → Buyer agreement signed → Closed, with an automation that opens a same-day task the moment anyone reaches "Replied" or later. operator [19]

Tags and custom fields

Set these up before you import a single contact, not after:

Tags: one per lead source, one per import batch, plus a standing suppressed tag.
Custom fields on every contact: source (where the lead came from), list_source (the exact list or campaign, if imported), registration_date or date added, has_agent and already_transacted (both suppress from outbound), consent_email (Y/N, with the provenance note and a timestamp of when and how it was given), and consent_sms (N until earned, then Y with the timestamp and exact method, for example "replied YES to quiz text 2026-09-14"). Never set a consent field to Y without a recorded date and method attached. operator [19]

Sources and lead routing

Every lead source (portal, sphere, open house, past client referral, website form) should write straight into the source field on arrival, not get hand-tagged later. Route by source: portal leads to speed-to-lead automation and a same-day human follow-up task; sphere and referral leads straight to a personal task, no automation; a walk-in or open house sign-in to the same-day task plus the nurture cadence for that lead type (Chapter 05 covers the cadences).

The daily "money list" view

Build one saved view, checked first every morning: contacts in an active pipeline stage (agreement signed through closed) with no task logged in the last 3 days, plus anyone who moved to "Replied / Engaged" or later in the last 24 hours. That view, not the full contact list, is what you work first each day.

Importing and cleaning a contact list

Build the suppression segment first, before importing anything else: prior opt-outs from any of your properties, hard bounces, "already bought," and "has agent." Wire the STOP/unsubscribe automation before a single campaign exists.
Add a provenance column to the import file itself: where the list came from, the date, and the specific form or event that produced it. Records missing provenance get tagged no-provenance and are excluded from every send until someone can confirm the source. Unknown origin defaults to no-send, not "probably fine." operator [19]
Dedupe on email and phone before import, keeping the most recent record when two rows conflict, and log how many duplicates were merged so you have a real count of the list size.
Only after suppression and provenance are wired: import the cleaned file, mapping source, list_source, and the date fields on the way in, not as a follow-up task.

Speed-to-lead automation done safely

Third-party tools built specifically for instant lead response, such as Structurely's automated SMS/email first-touch qualification, run $249 to $299/mo (Team about $499/mo), plus a roughly $2,000 one-time onboarding fee. report [25]

The safe version of "speed to lead" is narrower than most of these tools advertise on their own homepages: auto-reply by text only to a number that already carries a consent_sms = Y record with a timestamp and method. For everyone else, the automation's job is to get an email out fast and to open a same-day human task, not to text a number that never opted in. An auto-text sent to an unconsented number is not a growth hack, it is the exact TCPA exposure pattern described in Chapter 05 ($500 to $1,500 per text, no proof of harm required). Build the routing so the CRM checks the consent field before an SMS step ever fires, and route everyone else into an email-first sequence instead.

CRM connections to Claude: three routes, ranked

Copy, paste, and CSV export into a Claude Project. Lowest effort, lowest risk. Export a segment (this week's new leads, last week's activity) as a CSV, drop it into a Claude Project alongside your office file, and ask Claude to summarize, flag, or draft from it. Nothing writes back to the CRM automatically, so there is no accidental send or save. Claude Projects are available on all plans, with expanded knowledge capacity on paid tiers. verified [22]
Zapier MCP or Make MCP, as documented by those vendors. More automated, more moving parts. Zapier's hosted MCP server connects Claude to thousands of apps' existing Zapier actions; Make's official MCP server lets Claude run scenarios and manage a Make account. Both are the vendor's own documented product, not an Anthropic feature, and both can be wired to write into a CRM if you build it that way. claim [20][21] Use these for read-only lookups and drafts first; only add a write action once you trust the mapping.
Official vendor MCP servers, where they exist. HubSpot's Remote MCP Server (GA April 13, 2026), GoHighLevel's official MCP (about 39 tools, free on Unlimited/SaaS Pro), and Twenty's native MCP server are the current examples in this space with a vendor actually running and supporting the connector. verified [9][15][17] This is the most capable route and also the one that most directly touches production data.

The rule across all three routes, no exceptions: Claude reads and drafts, a human clicks save or send. Anthropic's own Usage Policy treats housing-adjacent decisions as high-risk, requiring a qualified professional to review output before it is finalized and disclosure of AI use to the affected person. verified [23] Treat any CRM write action (a new tag, a stage change, an outbound message) the same way: proposed by Claude, executed by you.

Claude workflow blocks

Clean a contact import with Claude

Trigger: you have a raw CSV to bring into the CRM, from a portal export, a past-client spreadsheet, or a purchased/transferred list.

Claude prepares: reads the CSV in a Project, flags likely duplicate rows (matching email or phone), flags rows missing a provenance value, standardizes phone and email formatting, and proposes a mapping from the file's columns to your CRM's source, list_source, consent_email, and consent_sms fields.

Human checks and does: reviews the flagged duplicates and no-provenance rows, decides what gets excluded rather than trusting the flag blindly, and performs the actual import into the CRM.

Never automated: Claude does not write directly into the CRM's live contact database from this step, and it never marks a consent field Y on Claude's own judgment; consent status comes only from a real, dated event.

The weekly relationship review

Trigger: once a week, same day and time, before your calling block.

Claude prepares: reads an export of the last week's CRM activity (calls, texts, stage changes, no-activity flags) and proposes ten people to call, with a one-line reason for each: a milestone date, a stalled stage, a past client's home-anniversary, a lead that went quiet after an initial reply.

Human checks and does: reads the ten names and reasons, adjusts the list against what Claude can't see (a personal note, a recent conversation elsewhere), and makes the calls.

Never automated: the calls themselves, and any promise or commitment made on one.

Draft the follow-up, human sends

Trigger: a lead or client needs a follow-up message after a call, a showing, or a stalled stage.

Claude prepares: a short, personalized draft using the CRM notes and the person's stage, in your voice, with no invented facts about the property or the market.

Human checks and does: reads it for accuracy, tone, and consent (is this person actually textable), and hits send from the CRM or their own phone.

Never automated: the send. No route in this chapter, including the official vendor MCP servers, should be wired to send a message without that human click.

What the CRM must never do by itself

Text or call a number that does not carry a dated, recorded consent event for that channel.
Mark a consent field Y without a real timestamp and method attached to it.
Send a bulk campaign to a list that has not been run through the suppression segment first.
Let an automation, AI feature, or MCP write action fire an outbound message without a human review step in front of it.
Mix an unconsented cold list into the same database as your working client relationships.

What stays human

Deciding whether a contact is legally textable, and recording the consent event that makes it true.
Clicking save on any CRM record change and send on any outbound message, regardless of which Claude route drafted it.
Choosing which CRM to buy and which tier, based on your own volume and budget, not a vendor's marketing page alone.
Reviewing the weekly relationship review list before calling anyone on it.

Do this today

Pick one CRM from the comparison table and set up the buyer and seller pipeline stages, before importing any contacts.
Add the source, consent_email, and consent_sms custom fields, with provenance and timestamp, to every contact record.
Export your current contact list to CSV, open a Claude Project, and run the "Clean a contact import" workflow before re-importing anything.

Sources

Follow Up Boss Pricing, followupboss.com/pricing, retrieved 2026-09-03
Follow Up Boss API docs, docs.followupboss.com, retrieved 2026-09-03
Follow Up Boss third-party MCP integrations, followupace.com/integrations, retrieved 2026-09-03
Lofty Pricing, lofty.com/pricing, retrieved 2026-09-03
Lofty developer API, developer.lofty.com, retrieved 2026-09-03
Lofty A2P 10DLC help article, help.lofty.com, retrieved 2026-09-03
kvCORE/BoldTrail pricing estimate, theprotoolkit.com, retrieved 2026-09-03
BoldTrail TCPA compliance article, help.insiderealestate.com, retrieved 2026-09-03
GoHighLevel Pricing, gohighlevel.com/pricing, retrieved 2026-09-03
GoHighLevel MCP help article, help.gohighlevel.com, retrieved 2026-09-03
GoHighLevel A2P registration guide, help.gohighlevel.com, retrieved 2026-09-03
Sierra Interactive Pricing, sierrainteractive.com/pricing, retrieved 2026-09-03
Sierra Interactive texting best practices, help.sierrainteractive.com, retrieved 2026-09-03
Wise Agent Pricing, wiseagent.com/pricing, retrieved 2026-09-03
HubSpot Free CRM pricing, hubspot.com/pricing/crm, retrieved 2026-09-03
HubSpot MCP documentation, developers.hubspot.com/ai-tools/mcp, retrieved 2026-09-03
Twenty Pricing, twenty.com/pricing, retrieved 2026-09-03
Notion Pricing, notion.com/pricing, retrieved 2026-09-03
NHF AZ CRM Stack: Lofty + GoHighLevel plan, operator procedure, 2026-08-19
Zapier MCP documentation, zapier.com/mcp, retrieved 2026-09-03
Make MCP server documentation, developers.make.com/mcp-server, retrieved 2026-09-03
Claude Projects support article, support.claude.com/en/articles/9517075, retrieved 2026-09-03
Anthropic Usage Policy, anthropic.com/aup, retrieved 2026-09-03
Notion AI/API summary, notion.com/pricing, retrieved 2026-09-03
GoHighLevel AI/MCP summary, gohighlevel.com/pricing, retrieved 2026-09-03
Structurely pricing comparison, followupace.com/compare/structurely, retrieved 2026-09-03

← Previous · All chapters · Next →

Starting from zero? Use the field manuals first.

02 · BEFORE YOU BUY SOFTWARE
If you're still choosing the buyer, offer, stack, agents, content, customers, and delivery, don't start with a tool: start with the manuals. These create local worksheets only: no checkout, outreach, or account setup.

Find the first manual

Not sure which guide to open first? Start from your stuck point and let the finder recommend the next manual and the first worksheet question. No personal data is collected: it just points you at the right starting move.

Open the guide finder →

Work the starter pack

Move through buyer, offer, stack, agents, content, customers, and delivery as a sequence of short local worksheet sessions. One artifact at a time, in order, so you build the business instead of browsing for it.

See the Business Path →

See what "done" looks like

Read a fully worked field manual before writing your own version, so the output feels concrete: a real, numbered workflow with the risky steps gated, not a fake proof claim.

Open a sample field manual →