Skip to main content

Developers · MCP Available

Build in Kinetic with AI through the MCP server

Kinetic's MCP server connects an AI client such as Claude Code or Cursor to a Kinetic space. Most teams use it to build: forms, Kapps, workflows and integrations. It can also create records and start workflows. It signs in as a Kinetic user and can do only what that user is permitted to do.

Book a working session Developer hub

Reach Every operation in the Kinetic Core and Integrator APIs, limited by the signed-in user's security policies.

Last updated
Status
Generally available

Scope

What AI can do through MCP

The server is generated from the Kinetic API specifications, so the AI client can make any call its user could. By default it groups operations into one tool per resource, each taking an action parameter.

Tasks an AI client can perform through the Kinetic MCP server, and the tools it uses
Task Tools Notes
Create and edit Kapps and forms kapps, forms Pages, fields, conditions and form types.
Create and edit workflows workflows Workflow definitions attached to a space, a Kapp or a form.
Set up integrations connections, operations Integrator Connections and the Operations that call them.
Create and update records submissions Create, update and search submissions.
Start a workflow submissions Submitting a form starts the workflow attached to it, as it would for a person.
Reach any other endpoint get_api_spec, execute_api Look up an operation in the bundled API specs, then call it.

The server also covers users, teams, webhooks and security policy definitions. A space administrator's account would let the AI change those too, so scope the account to the build.

Permissions come from the user it signs in as

The server has no access of its own. It signs in to one space with a Kinetic username and password, and every call is checked against that user's security policies, the same as a call from the console or the REST API.

One space per connection
Connect it to a development space. Changes reach production when a person promotes them.
Workflows run deterministically
When the AI submits a form, the workflow engine runs that form's workflow step by step, as it would for any other submission. The model decides what to submit. It does not decide what the workflow does next.
SSO-only accounts
The Integrator tools (Connections and Operations) need a local Kinetic account with a password. A user who signs in only through single sign-on cannot use them.
No bundled model
You bring the MCP client and the model behind it.

Setup

Connect an MCP client

The server is a Node.js process. Most clients start it over stdio; it can also serve HTTP for web-based clients.

  1. 1. Create the account. A user in your development space with access to the Kapps the AI will work on.
  2. 2. Put credentials in a .env file. Never in the client config, which is often committed or synced.
  3. 3. Register the server with your client. The example is a Claude Code .mcp.json; Cursor uses the same shape. Node.js 20.6 or later is needed for --env-file.

Installation steps are in the MCP documentation published with the release.

.env · space and credentials 
# .env  (keep it out of version control and out of the client config)KINETIC_SERVER_URL=https://<your-space-url>KINETIC_USERNAME=<ai-build-user>KINETIC_PASSWORD=<password> # Optional. consolidated is the default; slim registers the fewest tools.KINETIC_MCP_MODE=consolidated

Tested on version: pending verification

.mcp.json · Claude Code 
{  "mcpServers": {    "kinetic-platform": {      "command": "node",      "args": [        "--env-file=/<path-to-kinetic-platform-mcp>/.env",        "/<path-to-kinetic-platform-mcp>/dist/index.js",        "--stdio"      ]    }  }}

Tested on version: pending verification

A typical build session

  1. 01

    Describe

    Someone describes the form and process in the AI client and names the systems it touches.

  2. 02

    Read the space

    The AI lists the Kapps, forms and Operations it can see, so it builds on what exists.

  3. 03

    Build

    It creates or edits the definitions in a development space.

  4. 04

    Test

    It, or a person, submits test requests. Each submission starts the real workflow, and the run history shows every step.

  5. 05

    Review and promote

    A person checks the change in the Space Console, then exports, diffs and imports it into production.

How AI-built changes are reviewed

What the AI builds is ordinary Kinetic configuration, so it goes through the review tools your team already uses.

Where to go next

MCP questions

Answered by Kinetic Data. Last updated .

  • Is the MCP server available now?

    Yes. It is generally available. To get the server and setup instructions for your environment, book a working session or ask your Kinetic account team.

  • Can AI run workflows through MCP?

    It can start them. Creating a submission starts the workflow attached to that form, exactly as it would for a person. The workflow engine then runs the defined steps in order; the AI client does not choose or change the steps while a workflow runs.

  • What stops the AI from changing production?

    The account it signs in with. Give the AI its own user in a development space and promote changes to production by export and import.

  • Which AI clients and models can I use?

    Any client that supports MCP. The server ships with configuration examples for Claude Code, Claude Desktop and Cursor. Kinetic does not supply a model: use Anthropic Claude, Amazon Bedrock, Azure OpenAI, a self-hosted or open model, or any other model your MCP client supports.

  • Is there a Slack or Teams app?

    No. Kinetic is headless. Teams bring it into Slack, Teams or another chat tool by building a plugin against the Kinetic API, or by using an AI assistant that supports MCP.

  • How is MCP different from calling the REST API?

    It calls the same APIs under the same permissions. The AI client discovers operations from the tool descriptions and the bundled API specs, so nobody writes a wrapper per call.

Watch AI build one of your processes

Bring a process that crosses two or more systems. In a free two-hour working session we build it in a Kinetic space with an AI client, show you the definitions it produced, and leave you with a written plan.

Book a working session