The platform

Structure decides. Operators execute.

An engagement is a plan with a clock. It spans weeks, survives exceptions, and leaves a record of every step it took, across phone, text, email, and the systems you already run.

The working loop

Listen, decide, act, learn.

Goal: Clear an unreleased mortgage and a judgment lien before commitment, file 24-10877.

  • Planned flow
  • Path that went nowhere
  • New path or retry
LISTENwhat arrivedDECIDEwhich pathACTwork performedLEARNwhat it keepsDay 1two exceptionsDay 1open two tracksDay 1call servicerDay 1wrong servicer,path endsDay 2open a new pathDay 4request payoffand releaseDay 4watch the expirydateDay 10quote expiredDay 10re-request,attempt 3Day 12hand judgment toexaminerDay 14both exceptionscleared
On planDay 1 · listen lane · step 1 of 11

Search comes back with two problems

An unreleased 2019 mortgage and a docketed judgment against a similar name. Neither clears itself, and both have to be gone before commitment.

  • exception: open_mortgage_2019
  • exception: judgment_lien
  • party: prior servicer

The plan is deterministic. Conditions, waits, contact limits, and escalation paths are defined before an engagement starts, and they do not change because a conversation went sideways. AI drafts the words inside those rules, the sentence in a text, the summary in an email.

Between events, nothing runs. An operator wakes on a reply, a portal update, or a scheduled step, does the work, records the outcome, and goes quiet. That is what makes thousands of simultaneous open files a matter of structure rather than staffing.

One engagement

Walk a file, day by day.

Pick a workflow, then step through the days. Each point on the track is a signal the operator saw, a decision it made, or an action it took.

GoalCarry file 24-10877 from order to recorded.

Day 1 – Day 33
Day 1Day 33
SignalDay 18:12 am

New order received

Order lands from the production system with the property, parties, and a target closing date 32 days out.

Emitted

  • Acknowledge to sending agent
  • Build the engagement clock
SignalDecisionActionHandoff

The open runtime

Connect what you already run.

Documented API

Every operator capability, start an engagement, read its record, write back a result, is a documented endpoint you can call yourself.

OpenAPI connectors

Register a system with a published OpenAPI spec and operators can use it in engagements. Configuration, not an integration project.

MCP servers

Operators call any MCP server: TitleMCP for title data, your own tools, or third-party servers. Every call is typed and every response is schema-defined.

Decision replay

Reconstruct any engagement step by step: the inputs it saw, the rule it applied, the action it took, and the outcome it learned from.

Your systems

Categories, not a fixed list. If a system publishes an OpenAPI spec or an MCP server, an operator can use it.

  • Title production
  • E-recording
  • E-sign
  • Underwriter portal
  • Telephony
  • MCP servers
  • Payoff portals
  • Lien searches

TitleMCP is the open-source collection Razi invented and publishes for title work, but the runtime is not limited to it. Any MCP server your team runs or trusts can be wired in the same way.

TitleMCP

The protocol layer underneath

We built TitleMCP, and our operators speak it natively.

TitleMCP is an open-source collection of Model Context Protocol servers and jurisdiction-aware adapters for title, reusable primitives for estoppels, municipal liens, payoffs, tax certificates, releases, FinCEN reporting, and curative work. Razi Title, Inc. created it, released it under the MIT license, and runs on it in production. It is the title industry's only open tooling of its kind.

That matters here for one reason: when a Razi operator needs a parcel record or a payoff figure, it calls a typed tool with a defined schema instead of guessing at a vendor portal. The result lands on the file as data, auditable, replayable, and attributable to the exact call that produced it.

Open source
MIT licensed, public schemas
The only open protocol layer built for title work, anyone can read it, fork it, or ship an adapter.
We wrote it
Razi Title, Inc.
We created TitleMCP because our own operators needed a stable, typed way to reach county and vendor systems.
We run on it
Native, not a bolt-on
Razi operators call TitleMCP tools directly, so every lookup returns schema-typed data instead of scraped guesses.

OPERATOR · TOOL CALLS ON FILE 24-10877

Live
  • → tooltitlemcp-releases · watch_release

    title.releases.v1

    { mortgage: "2019-OR-44182" }

    ← returns typed ReleaseStatus

  • → tooltitlemcp-payoff · parse_letter

    title.payoff.v1

    { document_id: doc_8841 }

    ← returns typed PayoffRecord

  • → tooltitlemcp-liens · search_municipal

    title.liens.v1

    { address: "1412 Maple Ave" }

    ← returns typed LienSearchResult

  • → tooltitlemcp-curative · open_curative

    title.curative.v1

    { defect: "judgment_lien" }

    ← returns typed CurativeItem

Same schema every time, in every jurisdiction we support, which is why the decision trace behind a cleared exception can be read back months later.

Hire Razi

The same record that makes an engagement replayable is what makes an invoice auditable, so see how the platform fee and completion fee work, and how tenant isolation, consent, and the audit trail are handled.

Hire Razi