How to Set Up Knack’s MCP Server: A Step-by-Step Guide

Knack’s MCP server lets AI coding tools like Lovable, Bolt, Base44, and Claude Code connect directly to Knack as their backend. When you complete the knack mcp server setup, your AI tool reads your Knack schema and generates accurate frontend components. At runtime, all data operations route to Knack’s infrastructure via the API. For healthcare teams, this means a vibe-coded frontend backed by a HIPAA-compliant database from the first build session. Here is exactly how to set it up.

Key takeaways

• Knack’s MCP server URL is https://mcp.knack.com/mcp. It uses OAuth for authentication with no API keys or tokens required.
• The MCP connection is a build-time integration. Your AI tool uses it to read your Knack schema and generate accurate frontend components. At runtime, live data operations run through Knack’s API, not the MCP server.
• Supported tools currently include Lovable, Bolt, Base44, Claude Code, ChatGPT, and Replit.
• For healthcare teams handling PHI, a Knack Health HIPAA plan is required before any real patient data enters the system. The MCP setup process is identical on HIPAA plans.
• Build your data schema in Knack before connecting your AI tool. The MCP server reads your existing schema so it works best when there is already a schema to read.

What the knack mcp server setup actually does

Before walking through the steps, it helps to understand what the MCP connection does and does not do. Specifically, MCP stands for Model Context Protocol, an open standard for connecting AI coding tools to external services. When you add Knack’s MCP server to your AI tool’s configuration, the AI gains callable functions. These go beyond its built-in knowledge, so instead of describing how to integrate with Knack, the tool can just do it.

Specifically, the MCP server manages your Knack application’s data model. It gives the AI the ability to create tables, define fields, and set up relationships. App structure updates also happen through natural language conversation. All tables, fields, and data are created and stored inside your Knack application on Knack’s hosted infrastructure. Notably, nothing is stored locally or inside the AI tool itself.

Most importantly, the MCP connection is a build-time tool only. When your app is live and users are creating or reading records, those operations run through Knack’s runtime API instead. Most importantly, this distinction matters for healthcare teams. The compliance question is about where PHI lands at runtime, not which tool the builder uses during development. For a deeper look at how the build-time and runtime split works, the headless Knack guide covers the full architecture.

Before you start: what to prepare in Knack first

Consequently, the MCP server reads your existing Knack data schema. The integration therefore works best when you define your data structure in Knack before connecting your AI tool.

Create a Knack account or sign in

If you do not already have a Knack account, start a free trial at knack.com. For healthcare apps that will handle PHI, a Knack Health HIPAA plan is required before going live with real patient data. Most importantly, the MCP setup is identical on free and HIPAA plans. You can therefore complete the technical connection on a trial account first. Upgrade to a HIPAA plan when you are ready to go live with real patient data.

Build your data schema

In the Knack Builder, create the tables, fields, and relationships your app needs. For example, a patient intake app might have a Patients table with name, date of birth, and contact fields. An Appointments table connected to it would be a typical companion structure. In general, getting this structure right before building the frontend saves significant rework later. The AI generates components based on what it finds in your schema, so a more complete schema produces more accurate output.

Note your Knack Builder credentials

The MCP server authenticates via OAuth using your Knack Dashboard and Builder credentials. Specifically, you do not need API keys or developer tokens for the MCP connection — just your standard Knack login.

How to complete the knack mcp server setup

Overall, the core setup is the same regardless of which AI coding tool you are using. Add the MCP server URL to your tool’s MCP configuration, sign in with your Knack credentials, and the connection is live.

Step 1: Add the MCP server URL

In your AI coding tool, navigate to the MCP or integrations configuration area. Add the following URL:

https://mcp.knack.com/mcp

The exact location of this setting varies by tool. Generally, look for an MCP servers, integrations, or backend configuration panel. For the exact configuration path in each supported tool, the full documentation at docs.knack.com/docs/mcp covers tool-specific steps.

Step 2: Authenticate with your Knack credentials

After adding the MCP server URL, your AI tool will prompt you to authenticate. Sign in using your Knack Builder credentials through the standard OAuth flow. No API keys or manual token management are required. Once authenticated, the AI tool reads your Knack application’s schema, including your table names, field names, data types, and relationships.

Step 3: Confirm the connection

After authenticating, ask your AI tool a question about your Knack schema to confirm it has read the data model correctly. For example, ask it to list the tables in your Knack application. Alternatively, ask it to describe the fields in a specific table. If it returns accurate schema information, the MCP connection is working. At that point, you are ready to start building.

Step 4: Build your frontend

With the MCP server connected, your AI tool understands your Knack data model. From this point, describe what you need to build in plain language. Consequently, the AI generates frontend components, forms, and UI elements that map directly to your actual Knack fields and relationships. Generic placeholders that require manual reconciliation afterward are no longer the starting point. Instead, the output is already aligned to your real data model.

Here is a concrete example of the kind of prompt that works well. Tell your AI tool: “Build a patient intake form that collects name, date of birth, phone number, and reason for visit. Save each submission as a new record in the Patients table.” The AI then generates the form. It also creates the corresponding Knack API calls using the field names and table structure it read from your schema.

Step 5: Confirm your BAA before going live with PHI

Before any real patient data enters the system, confirm your BAA is executed with the Knack Health team. The BAA covers all PHI stored in Knack Health’s infrastructure. Only after your BAA is in place should real patient data flow through the app. For more detail on what a business associate agreement covers and why it is required, the BAA explainer covers the specifics.

How runtime data flow works after setup

Once your app is live, the MCP server is no longer active. Instead, your frontend communicates with Knack via Knack’s runtime API. Every record operation goes through this API connection. That includes creating a patient record, reading an intake submission, or updating an appointment status.

For healthcare teams, this architecture means PHI flows directly between the user’s browser and Knack Health’s HIPAA-compliant backend. Specifically, the AI coding tool’s infrastructure handles the frontend interface and is not in the data path. Consequently, the frontend tool does not need to be HIPAA compliant for the overall system to be compliant. The compliance boundary sits at the Knack layer. Specifically, that includes the signed BAA, encryption at rest and in transit, role-based access controls, and record change logs.

This is why connecting Lovable, Bolt, Base44, or Claude Code to Knack Health via MCP produces a production-ready HIPAA architecture. For a broader overview of secure vibe coding across all supported tools, that guide covers each step and its compliance implications.

Tool-specific configuration notes

Lovable

In Lovable, open your project settings and find the MCP servers section. Add the URL https://mcp.knack.com/mcp and authenticate. Once connected, Lovable’s AI builder generates frontend components that map to your Knack schema. If you have an existing Lovable prototype with data, the Knack Health team can help migrate your data models and records. You do not need to rebuild your frontend to make the switch. The full Lovable and Knack Health setup guide covers the migration process in detail.

Bolt

In Bolt, navigate to the integrations or backend configuration panel and add the Knack MCP server URL. Once connected, Bolt’s AI generates components against your Knack schema. Notably, Bolt’s default database backend does not include a BAA. The MCP connection to Knack Health replaces that default data layer with HIPAA-compliant infrastructure. For more detail on why Bolt is not HIPAA compliant on its own, that post explains the gap. It also covers how the Knack connection closes it.

Base44

In Base44, navigate to the backend or integrations settings and add the Knack MCP server URL. The process is equivalent to Lovable. For teams migrating an existing Base44 prototype, the MCP connection lets you keep the frontend and switch the backend. More detail is in the Base44 and Knack Health guide.

Claude Code

In Claude Code, add the Knack MCP server to your MCP configuration file. Once configured, Claude Code reads your Knack schema and generates application code that references your actual Knack data structure. Forms and API calls are included in that output. The Claude Code and Knack Health setup guide covers the configuration file format and exact steps.

What Knack Health provides on a HIPAA plan

For healthcare teams, the value of the knack mcp server setup is not just the development workflow. Specifically, it is what runs behind the connection on a Knack Health HIPAA plan.

Signed BAA

Every Knack Health HIPAA plan includes a signed Business Associate Agreement. In particular, it covers all PHI stored in Knack’s infrastructure. That includes data flowing in from any AI-built frontend connected through the MCP server.

Encryption at rest and in transit

All PHI stored in Knack Health is encrypted at rest. Furthermore, all data transmitted between the frontend, the runtime API, and Knack’s backend is encrypted in transit.

Role-based access controls

Knack Health enforces field-level permissions across every view, form, and API call. For example, a front desk coordinator sees scheduling fields. Clinical staff see clinical notes, while billing staff see financial data only.

Record change logs

Every access to and change of a patient record is logged automatically. The log captures who performed the action, when, and what values changed. Consequently, this satisfies HIPAA’s record-keeping requirements without additional configuration.

HIPAA-ready infrastructure

Knack Health runs on AWS GovCloud hosting, the same infrastructure tier used by US government agencies with strict data security requirements.

Knack Health’s entry-level HIPAA plan starts at $499 per month with no per-user fees. Advanced plans are available with expanded features for larger healthcare organizations. For full details, visit the Knack Health product page or the AI app builder page.