Scandit MCP Installation Guide

Connect the Scandit MCP server to your AI coding agent and go from an idea to a working scanner in one conversation. You only need to authenticate in the browser. You do not hunt through the dashboard or paste a key by hand: the server picks the product, provisions a trial key, and fetches it into your project.

It works with MCP-compatible clients that support streamable HTTP and browser OAuth, including Claude Code, Cursor, and VS Code.

Connecting is free, and so are the keys it creates. The server provisions trial license keys and can read the ones you already have. It never creates, revokes, or modifies production licenses.

Server URL

https://ssl.scandit.com/mcp

The URL must be configured exactly as shown, ending in /mcp.

Install

Installed the Scandit plugin?

The Scandit plugin already includes this server when you install it with npx plugins add scandit/skills or from the Claude Code, Cursor, Codex CLI, or Copilot CLI marketplace. Skip this section and go to Authentication. Adding the server here as well gives you two Scandit servers, and each one needs its own sign-in.

Claude Code

claude mcp add --transport http scandit https://ssl.scandit.com/mcp

Then start a fresh claude session in your project directory.

Cursor

Add the server to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "scandit": { "url": "https://ssl.scandit.com/mcp" }
  }
}

VS Code

Add the server to your MCP user configuration (Command Palette → MCP: Open User Configuration):

{
  "servers": {
    "scandit": {
      "type": "http",
      "url": "https://ssl.scandit.com/mcp"
    }
  }
}

Any other client

Point your client at https://ssl.scandit.com/mcp over streamable HTTP (for example, Codex or Zed). No API key or local process is required. The client must be able to open a browser and complete OAuth with an HTTP loopback redirect (http://localhost or http://127.0.0.1).

Authentication

Authentication is browser-based (OAuth 2.1 with PKCE), and you start it yourself. If a Scandit tool is called before you have authenticated, the agent asks you to authenticate rather than opening a browser.

• Claude Code: run /mcp, select scandit (plugin:scandit-sdk:scandit if the server came with the plugin), and authenticate.
• Cursor: go to Settings → Tools & MCP → scandit → Connect / Authenticate.
• Other clients: use the client’s own connect or authenticate control for the server.

The tab lands on sign-up; if you already have a Scandit account, use the Sign in link on that page. After you approve access, tell the agent to continue.

Signing in does not create a subscription or a license key. The agent still has to provision one, which it does on its next step.

Limitation: Headless and CI authentication is not supported yet.

Disconnect

Remove or disconnect the Scandit server in the client you used to connect. That drops the tokens the agent stored, so it can no longer call Scandit tools.

• Claude Code: run claude mcp remove scandit. If the server came with the plugin, disable or uninstall the scandit-sdk plugin in /plugin instead. This also removes the skills.
• Cursor: disconnect or remove scandit in Settings → Tools & MCP.
• VS Code: remove the scandit entry from your MCP user configuration.

Signing out of the dashboard does not disconnect the agent. License keys and the trial stay in the dashboard; delete them there if you no longer want them. There is no dashboard page yet to revoke MCP access on the server.

Get started

Install the server and connect it, then in an empty project directory prompt your agent:

Build me a barcode scanner .NET iOS app with Scandit.

1. The agent works out which Scandit product fits your use case and which framework you are building with, asking you if either is unclear.
2. It calls ensure_scanner_setup, which creates or reuses a matching trial subscription, project, and license key for your platform.
3. It writes the key into your project’s .env file. For native project configuration that a .env file cannot hold, it retrieves the key with get_license_key.
4. It follows the matching Agent Skill to scaffold and run the app.

The MCP server handles the account and license key; the Agent Skills handle the SDK integration itself. For the best result, install the Scandit plugin, which includes both the skills and this server:

npx plugins add scandit/skills

If you already added the server with the Install steps, remove that copy first (in Claude Code, claude mcp remove scandit) so you do not end up with two Scandit servers.

What you can do

• Create your account and a trial through the browser, with no forms to fill in the chat.
• Provision a trial license key for the product and platform you are building for, and write it to .env.
• List the license keys you already have, across every subscription your account can access.
• Add another key to an existing trial for a second app identifier, another platform, or an extra web domain.
• Retrieve a key to paste into native project configuration.

Keys created this way appear in the Scandit dashboard under License Keys, labeled Created via MCP, alongside keys you create yourself.

Tools reference

Tool	What it does
ensure_scanner_setup	Creates or reuses the trial subscription, project, and license key for the chosen product and platforms, and returns a command that writes the key to .env. Idempotent: safe to call again.
create_trial_license_key	Adds or reuses an additional key on an existing trial, for another app identifier, platform, or web domain.
list_license_keys	Lists keys across all accessible subscriptions as masked previews. Never returns a usable key.
get_license_key	Returns one full license key by ID, for configuration a .env file cannot hold.
get_license_env_command	Returns a one-line command that writes a key into a .env file without exposing it in the chat.
whoami	Returns the signed-in account’s email, account ID, and default subscription.

Claude Code users can also run /mcp__scandit__get-started for an orientation prompt, or /mcp__plugin_scandit-sdk_scandit__get-started if the server came with the plugin.

Products and platforms

ensure_scanner_setup takes the product your agent selected (sdk, native, express, id-bolt, enterprise-browser, or wedge) plus the license platforms for your framework.

Framework	License platforms
Web	webassembly
iOS	ios
Android	android
.NET iOS	ios
.NET Android	android
Kotlin Multiplatform	ios, android
React Native, Flutter, Capacitor, Cordova, .NET MAUI	ios, android

If you already have an active trial, it is reused rather than replaced. You will not end up with duplicate trials for sdk, native, express, or id-bolt.

Web domains

For web keys, pass your production domain (for example, example.com). WebSDK trial keys already work on localhost, so there is no need to list it. One key can cover several domains. Where applicable, www.<domain> is added automatically.

Security

• License keys are secrets. The agent is instructed to show only a masked preview (for example, ATI3Bz…) in chat and to write the full key to .env using the returned command rather than pasting it into a file editor.
• list_license_keys never returns a usable key. Retrieving a full key is always an explicit, separate step.
• Keys your dashboard hides are never exposed in full, previewed, or reused for provisioning.
• Trials only. The server can create trial keys and read existing ones. It cannot create, revoke, or modify production licenses.
• .env writes stay inside your project. Only project-relative .env paths are accepted. Absolute paths, ~, and .. are rejected.
• Review tool calls. As with any MCP server, an agent can be steered by untrusted content in your project or on the web. Review what it does before approving.

FAQ

Do I need a Scandit account first?

No. The sign-in tab lets you create one, and the agent provisions a trial from there.

Does this cost anything?

You can try all Scandit products free for 30 days, no credit card required. Using Scandit beyond the trial in a production environment requires a commercial license.

Can I use it in CI or on a headless machine?

Not yet. Authentication currently requires a browser.

Can it touch my production license keys?

It can list them (masked) and, if you (or the agent) pick one, copy the full key into the project the same way the dashboard copy button would. It cannot create, revoke, or change those licenses. Keys the dashboard hides or marks non-copyable are never returned.

Where do the keys it creates show up?

In the dashboard under License Keys, with a Created via MCP label. New trial key projects created this way are named MCP Trial Keys.

My agent does not seem to know about Scandit.

In Claude Code, run /mcp__scandit__get-started, or /mcp__plugin_scandit-sdk_scandit__get-started if the server came with the plugin. If you have not installed the Agent Skills yet, install the Scandit plugin with npx plugins add scandit/skills.

No browser tab opened when the agent tried to use a tool.

Expected. You connect the server yourself. In Claude Code run /mcp; in Cursor use Settings → Tools & MCP → Connect. Then ask the agent to continue.

The consent screen names a localhost address.

That is expected. After you approve, the authorization code is sent to a loopback listener the client started on your machine (http://localhost:<port>/…). If you just connected this server in Claude Code, Cursor, or VS Code, that address is yours. If you did not start a connect flow, do not continue.

How do I disconnect?

Remove the Scandit server in the client: in Claude Code, run claude mcp remove scandit (if the server came with the plugin, disable or uninstall the scandit-sdk plugin in /plugin); in Cursor, go to Settings → Tools & MCP; in VS Code, remove it from your MCP configuration. Keys stay in the dashboard until you delete them.

Learn more

• Agent Skills: skills that write the SDK integration once a key is in place.
• scandit/skills repository: full catalog of Scandit skills.
• Model Context Protocol: the open spec behind MCP.

Need help? Contact support@scandit.com.