Real Estate AI Agent Runtime, APIs & MCP | Leni

Chatbots can generate answers. Production real-estate workflows are more demanding. They need identity, approved data, business definitions, context, tools, and a reliable way to finish work that takes longer than a single response. Leni provides that operating layer through an HTTP API for server applications and remote MCP for compatible AI assistants.

More than a prompt interface

General-purpose models are powerful engines for reasoning and drafting. Production real-estate workflows also need data access, business definitions, permissions, durable execution, context, and a clear boundary around what the agent is allowed to do.

Leni supplies that system around the model. Developers can build server-to-server applications with the HTTP API. Compatible assistants can connect through Leni’s remote MCP runtime. Both paths use Leni’s governed data and context services, subject to their own authentication method, permissions, plan, and usage rules.

What the runtime coordinates

Authorized real-estate data

Approved workflows can use permitted portfolio data through Leni’s semantic layer. Direct public MCP portfolio tools are read-only, and Leni enforces organization and property scope.

Universal Data Model and semantics

The runtime works with Leni’s normalized data foundation and governed business definitions so agents can reason about supported real-estate concepts consistently.

Personal and organizational context

A workflow can consult relevant personal context and, when the user’s plan and permissions allow, organization context. Context remains tied to the authenticated user and active organization.

Analysis and research

Leni supports real-estate analysis and market-research workflows through the governed runtime. Exact capabilities, inputs, and output contracts are defined in the current developer documentation.

Durable memory

Memory is available as an explicit capability. Durable memory is created, changed, or deleted only when the user asks and the authorization permits the action.

Durable execution

Longer analyses can be accepted asynchronously. A server-to-server client may receive HTTP 202 with stable run, message, and session identifiers, then poll for the result instead of resubmitting the work.

Two ways to build

HTTP API for server-to-server applications

Use the HTTP API when your application, backend, or internal service needs programmatic Leni access.

• Authenticate with a Leni project API key

• Keep the key on the server

• Follow the current request, upload, limit, usage, and polling contracts

• Use stable run and session identifiers for asynchronous work

• Do not place a project API key in an AI assistant or client-side application

The authoritative reference is the Leni developer documentation.

Remote MCP for compatible AI assistants

Use remote MCP when a compatible host needs to discover and call permitted Leni tools.

• MCP endpoint: https://mcp.leni.co/mcp

• Transport: remote Streamable HTTP

• Authentication: Leni OAuth sign-in and consent

• Scope: tied to the authenticated Leni user, active organization, plan, and permissions

For Claude, follow the current connector instructions in the Leni developer documentation. ChatGPT and other hosts can differ by product, plan, admin setting, and connector workflow; use the current documentation for each host rather than copying an old setup path.

Model-agnostic by design

Leni should not be framed as a replacement for Claude, ChatGPT, Gemini, or another model. The model is one part of the system. Leni provides the governed real-estate data, context, workflow, and delivery layer around the intelligence selected for the task.

That separation lets a team evolve its model choices without redefining every property connection, business rule, permission, and recurring workflow.

Clear capability boundaries

The public runtime should not be presented as an unrestricted operator of source systems. Current public boundaries do not support general claims of:

• Money movement

• Invoice payment

• Trading

• Regulated advice

• Mutation of property-management records

• Access outside the authenticated organization or authorized properties

• Raw SQL supplied by an MCP client

These constraints are part of making the runtime dependable for high-stakes work.

What teams can build

• A portfolio analysis assistant grounded in authorized property data

• A month-end reporting workflow that runs asynchronously

• A real-estate research application with source-linked outputs

• An internal assistant that applies governed definitions and organization context

• A developer product that uses Leni as its real-estate intelligence backend

• A Claude workflow that calls permitted Leni tools through remote MCP

Availability depends on implementation, plan, connected data, and user permissions.

Frequently asked questions

When do I need an agent runtime instead of only an LLM API?

An LLM API gives you model output. A production real-estate workflow also needs identity, permissions, tools, governed data, context, memory controls, and reliable execution. Leni brings those pieces together so your team does not have to rebuild the operating layer around every model or use case.

Is Leni an AI model?

Leni is the governed infrastructure around the model: real-estate data, context, analysis, tools, and runtime. We can use the appropriate model intelligence for a workflow without making your data and business logic depend on one model provider.

Should our developers use the Leni API or MCP?

Use the HTTP API for server-to-server products, back-end services, uploads, sessions, and asynchronous analysis. Use remote MCP when a compatible assistant needs to call Leni tools on behalf of a signed-in user. The developer documentation explains the current setup for each path.

Can I connect Leni to ChatGPT or Claude?

Yes, when your ChatGPT or Claude environment supports a remote MCP connection. Each user signs in with Leni OAuth, reviews the requested access, and works within their authorized organization and property scope.

Will Leni lock us into one model?

No. We designed Leni as model-agnostic infrastructure so the data, context, permissions, and workflow layer can remain consistent even when the right model changes.

Can I use Leni with Gemini?

Leni is model-agnostic, but we do not publish a direct Gemini connector unless both products support and document that path. If Gemini is part of your architecture, talk with us and we’ll confirm the supported integration route for your environment.

Build real-estate AI on a governed runtime

Read the developer documentation, connect through Leni MCP, or talk to Leni about your application and data scope.