Business MCP Layers
Connect AI to your business records and workflows
Build Lean develops custom MCP servers and integrations that let an AI assistant retrieve information and use defined tools across your business systems. We scope the records, permissions, and permitted actions around the questions and workflows your team needs to handle.
What is a business MCP layer?
MCP stands for Model Context Protocol. It gives AI applications a standard way to discover and call tools, such as looking up a work order, retrieving an invoice, or searching a document collection.
A business MCP layer is our term for the MCP servers, integrations, and access controls assembled around your company’s systems. Build Lean can configure suitable existing connectors, develop custom servers, and implement the business logic needed for your use case.
The AI application sends a structured request to an exposed tool. The server validates the request, checks access, and calls the connected system. It returns records or an action result for the AI application to use.
Person
Asks a question or requests a task
AI Application
Interprets the request and selects a tool
MCP Servers and Tools
Run defined operations on your systems
Access checks and validationConnected Systems
CRM, accounting, project tools, documents
Person
Asks a question or requests a task
AI Application
Interprets the request and selects a tool
MCP Servers and Tools
Run defined operations on your systems
Access checks and validationWhen updates are enabled: Approval UpdateConnected Systems
CRM, accounting, project tools, documents
Records, references, or action result return through the MCP server to the AI application.
AI Application
The assistant your team uses interprets the request and can select a relevant tool. Its connection settings and supported features determine how it works with the MCP servers.
MCP Servers and Tools
Tools define specific operations, their required inputs, and the results they return. Examples include finding an invoice by ID or listing open work orders for a property.
Connected Systems
Your CRM, accounting software, project tools, documents, and databases remain the sources of business information. Available APIs, permissions, and licensing determine what each integration can access or change.
An MCP layer can work directly with source systems or expose prepared data from a data lakehouse. A broader operations platform can add interfaces, approvals, and workflows around those connections.
Technical details
- Protocol
- MCP messages use JSON-RPC 2.0 between the AI application and each server.
- Transport
- Local servers typically run over standard input and output; remote servers use HTTP.
- Authorization
- Remote servers can use OAuth-based authorization, with credentials and scopes defined for each connected system.
- Optional retrieval
- Vector search can be added where a use case needs search across large document collections. It is optional and not part of MCP itself.
What a business MCP layer gives you
Give your team a practical way to retrieve business information and use defined actions through an AI assistant.
Less copying between systems
Let an assistant retrieve the records needed for a question, reducing the exports, pasted documents, and repeated lookups involved in everyday work.
Tools shaped around your operations
Expose useful tasks such as finding a project’s open RFIs or retrieving a customer’s outstanding invoices. Define the required fields, matching rules, and allowed actions around how your business works.
Answers you can check
Return record IDs, document references, and relevant timestamps alongside retrieved information. Give people a path back to the source when reviewing an AI-generated summary.
Defined access and operating responsibilities
Specify who can use each tool, which records it can reach, and which updates require approval. Document hosting, credentials, maintenance, and handover arrangements as part of the build.
How MCP could support your industry
Illustrative uses of custom MCP tools. Each build starts with the records your team needs, the systems that hold them, and the actions people should be allowed to take.
Look up project records and prepare the next step.
Retrieve RFIs, change orders, project details, and relevant documents using agreed project identifiers. An assistant could summarize open items with source references and draft a follow-up task for review. Creating or updating a task would use a separately authorized tool.
Available records, client compatibility, permissions, and update capabilities are assessed for each project.
How we build your MCP layer
Start with a specific business question or workflow. Define the tools, access, and expected behavior before expanding to more systems.
- 01 / 06
Map the Request and Source Records
Identify the questions people need answered, the records involved, and the systems that hold them. Define how projects, customers, properties, or matters are matched across sources.
- 02 / 06
Assess Connectors and Build the Missing Tools
Review existing MCP servers, connectors, APIs, and database interfaces. Configure suitable components and develop custom tools where needed, with clear input fields, output formats, and limits.
- 03 / 06
Configure Identity and Access
Define how people authenticate to the MCP layer and how it connects to each source system. Implement the required record and action restrictions, including additional checks where a shared service account is used.
- 04 / 06
Separate Retrieval From Updates
Define which tools only read information and which can change records. For updates, specify validation rules, approval requirements, and how the system handles repeated requests or records that have changed since review.
- 05 / 06
Test Answers, Actions, and Exceptions
Test the selected AI client against representative questions and expected results. Check source references, denied access, missing records, unavailable systems, and failed updates. Confirm what the user sees when a request cannot be completed.
- 06 / 06
Deploy and Define Support
Document hosting, configuration, tool usage, and operating responsibilities. Agree how credentials, API changes, logs, and failures will be maintained after launch.
Technical detail
MCP standardizes communication between the AI application and the server. The implementation still needs connections to source systems, business rules, and access controls. Server access and downstream system access may use different authorization mechanisms. Search indexes, caching, and additional workflow services are included where needed.
Illustrative Workflow
From an RFI question to an approved follow-up task
- 1Read
Ask About a Project
A project manager asks: “Which RFIs are overdue on Project J-1042?”
The assistant proposes a request to a tool that lists RFIs for a specified project. The tool checks the requester’s project access before retrieving records.
- 2Read
Apply the Agreed Rules
The tool evaluates due dates and status against the reporting date. It returns the RFIs that meet the agreed definition of overdue, along with record IDs, source links, and retrieval time.
If the source is unavailable, it returns an error. An unsuccessful lookup is not presented as “no overdue RFIs.”
- 3AI draft
Review the Answer and Proposed Action
The assistant summarizes the returned records and their references. The project manager can ask it to draft a follow-up task, then review the proposed owner, due date, and linked RFI.
The AI generates the wording. Application rules determine whether the requested action is permitted.
- 4ApprovalWrite
Approve and Confirm the Update
If task creation is included in the build, the authorized user approves the task details. The tool validates the request, checks for a duplicate, and submits it to the project system.
The system returns the created task ID or a failure status. The assistant reports that result without treating a draft as a completed update.
Simplified tool-call example
Shows the tool name and arguments. Protocol metadata and authentication details are omitted.
{
"name": "list_rfis",
"arguments": { "project_id": "J-1042", "status": "overdue" }
}The tool returns matching RFI IDs, due dates, status, source links, and the retrieval time, or an error if the project system cannot be reached.
Start with one useful connection.
Choose a recurring question that sends your team searching across systems, or a workflow that requires repeated copying and updates. We assess the available records and interfaces, then scope a first release around a defined set of tools.
The scope identifies which components will be configured, integrated, or custom-built. It also sets out access, testing, deployment, documentation, and ongoing operating responsibilities.
A defined implementation can be delivered as a Custom Software project. An AI Transformation partnership can support ongoing priorities, rollout, and adoption.
Business MCP layer questions
What owners and operators ask before connecting AI to their systems.
What should AI be able to look up or do for your team?
Tell us which question or workflow you want to improve and where the information lives today. We’ll discuss the systems involved, the access required, and a practical first connection.
Free · 30 minutes · No preparation required