AI Integration

Mobiscroll provides a set of machine-readable documentation files, AI behavior rules, and a live MCP server that enable coding assistants to generate accurate Connect API code. These prevent common AI issues like hallucinated endpoints, mixing Connect API calls with UI component code, and outdated authentication patterns.

Why AI integration?

AI coding assistants work best when they have access to structured, authoritative documentation. Without it, they often:

• Hallucinate APIs — invent endpoint paths, request parameters, response shapes, or SDK method signatures that don't exist
• Mix UI with backend — generate JSX component code when asked about the Connect REST API, or vice versa
• Reference outdated versions — generate API calls or SDK code that no longer match the current Connect schema
• Ignore authentication requirements — skip OAuth flows or use incorrect scopes for calendar access

The Mobiscroll AI integration solves these problems by providing Connect-specific documentation optimized for AI consumption, combined with behavior rules that enforce domain isolation, and an optional MCP server that serves live, version-stamped endpoint and SDK schemas on demand.

Architecture overview

The integration consists of four layers:

Data layer — llms files

Machine-readable documentation files containing the complete Mobiscroll Connect API reference and guides. These are the source of truth that AI tools read to answer questions.

File	Description
llms-connect-full.txt	Complete Mobiscroll Connect documentation
llms-connect.txt	Connect table of contents (links to individual pages)
llms-icons.txt	Icon names (IcoMoon, Font Awesome, Ionicons) — all frameworks

info

You don't need to download or host these files — the rules and routing layers reference them directly and fetch their content automatically.

Rules layer — .mdc files

Behavior rule files that tell AI assistants which package to use, which APIs are available, and what to avoid.

File	Domain
mobiscroll-connect.mdc	Mobiscroll Connect (backend / API)

Routing layer — CLAUDE.md

A context file specifically for Claude Code that provides domain detection signals, deterministic routing rules, API intent mapping, and anti-pattern examples. It ensures Claude selects the Connect documentation and never conflates Connect API calls with UI component code.

Live schema layer — MCP server

The Mobiscroll MCP server serves structured, version-stamped knowledge over the Model Context Protocol. It is a single, unified server: the same mobiscroll server that serves UI component schemas also exposes the Connect tools, generated directly from the Connect REST source and the 7-language SDK suite. The Connect tools are all prefixed Connect so they never collide with the UI component tools. Instead of relying on documentation snapshots, an assistant can call these tools to fetch the exact endpoint schema, SDK method signature, or cross-language equivalent it needs at generation time.

It is a hosted HTTP server at https://mcp.dev.mobiscroll.com/ — no local install required.

Tool	What it does
resolveConnectEnvironment	Detects which Connect SDK (language + version) a project uses from its dependency manifest, and echoes the served versions. Call this first.
listConnectEndpoints	Lists every Connect REST endpoint with its method, path, summary, and authentication.
getConnectEndpointSchema	Returns one endpoint's full schema — query/body params with types, authentication, responses, status codes, and examples.
listConnectSdkMethods	Lists a language's resources (auth / calendars / events) and the methods on each.
getConnectSdkMethod	Returns one SDK method's full signature, doc, params, return type, and example, in the language you pick.
searchConnect	Searches REST endpoints and SDK methods across all languages by keyword, ranked by relevance.
mapConnectEndpointToSdk	Maps a REST endpoint to its equivalent SDK call in each language — built on the SDKs' shared surface.
getConnectErrorTaxonomy	Returns the shared error categories and the idiomatic exception type for each language.

Which tool uses which files?

AI Tool	Documentation Source	Behavior Rules	Routing	Live lookups (MCP)
Cursor	llms-connect-full.txt via @docs	mobiscroll-connect.mdc	—	mobiscroll server (Connect tools)
GitHub Copilot	.mdc file (contains doc URLs)	.mdc file	—	mobiscroll server (Connect tools)
Claude Code	llms-connect-full.txt	CLAUDE.md	CLAUDE.md	mobiscroll server (Connect tools)

Cursor setup

Step 1: Register documentation sources

Open Cursor Settings → Indexing & Docs and add the documentation source for Connect:

Framework	Name	URL
Connect	Mobiscroll Connect	https://dev.mobiscroll.com/docs/llms-connect-full.txt

Only register the source matching your use case. Do not register multiple sources — this prevents cross-domain contamination.

Step 2: Add the rules file

Download the mobiscroll-connect.mdc file and place it in your project's .cursor/rules/ directory:

your-project/
├── .cursor/
│   └── rules/
│       └── mobiscroll-connect.mdc
├── src/
├── package.json
└── ...

Step 3: Configure the MCP server (Optional)

For live endpoint and SDK lookups, configure the Mobiscroll Connect MCP server so Cursor can call it during generation.

Create or edit .cursor/mcp.json in your project root:

{
  "mcpServers": {
    "mobiscroll": {
      "url": "https://mcp.dev.mobiscroll.com/"
    }
  }
}

No type field

Cursor infers the transport type from the URL. Do not add "type": "http" to Cursor's config — it causes an error.

your-project/
├── .cursor/
│   ├── mcp.json
│   └── rules/
│       └── mobiscroll-connect.mdc
├── src/
└── package.json

Scope	Config file	Shared with team
project	.cursor/mcp.json in project root	Yes, if committed
global	~/.cursor/mcp.json	No, all your projects

Verify the connection: Open the Output panel in Cursor and select MCP Logs from the dropdown. A successful connection logs tool discovery messages for the mobiscroll server, including the Connect-prefixed tools.

Step 4: Use @docs in queries

When asking Cursor about Mobiscroll Connect, include @docs to ensure it reads the registered documentation:

@docs How do I authenticate a user with the Mobiscroll Connect OAuth flow?

@docs How do I list all calendars for a connected Google account?

GitHub Copilot setup

Step 1: Add the rules file

Download the mobiscroll-connect.mdc file and place it at the root of your project or alternatively copy it's content to the rules files under the .github/ directory:

your-project/
├── mobiscroll-connect.mdc
├── .github/
|   ├── copilot-instructions.md       <-- Global repo rules
|   └── instructions/
|       └── connect-logic.instructions.md <-- Specific rules
├── src/
├── package.json
└── ...

The .mdc file will automatically influence Copilot responses when you work on files in that project. It tells Copilot which APIs are available and how to use them correctly.

Step 2: Configure the MCP server (Optional)

For live endpoint and SDK lookups, configure the Mobiscroll Connect MCP server so VS Code can call it during generation.

Create or edit .vscode/mcp.json in your project root:

{
  "servers": {
    "mobiscroll": {
      "type": "http",
      "url": "https://mcp.dev.mobiscroll.com/"
    }
  }
}

"servers" not "mcpServers"

VS Code uses "servers" as the root key — not "mcpServers" like Claude Code and Cursor. Using the wrong key silently breaks the config with no error message.

your-project/
├── .vscode/
│   └── mcp.json
├── .github/
│   └── instructions/
│       └── connect-logic.instructions.md
├── src/
└── package.json

Scope	Config file	Shared with team
workspace	.vscode/mcp.json in project root	Yes, if committed
user profile	Opened via MCP: Open User Configuration	No, all your workspaces

Verify the connection: Open the Command Palette and run MCP: List Servers. The mobiscroll server should appear with a connected status, exposing the Connect-prefixed tools. A trust dialog appears on first use — approve it to allow the server to start.

How it works

The .mdc file contains:

• Documentation URLs — points Copilot to the correct Connect docs
• Component mapping — maps user intents to the correct Mobiscroll Connect APIs
• Rules — enforces correct API usage, authentication flows, and webhook handling
• Constraints — prevents conflation of Connect REST endpoints with UI component APIs

With the MCP server configured, Copilot can additionally call the Connect tools for live endpoint and SDK schema lookups instead of relying on the documentation snapshot alone.

Claude Code setup

Set Connect up with config files: a CLAUDE.md routing file (Step 1) and the Mobiscroll MCP server (Step 2). Don't install the mobiscroll@mobiscroll plugin for Connect — its skills are currently UI-specific.

Step 1: Add CLAUDE.md

If you don't already have a CLAUDE.md in your project root, download CLAUDE.md and place it there. If you already have one, copy the contents into your existing file instead — see File contents below.

your-project/
├── CLAUDE.md
├── src/
├── package.json
└── ...

When Claude Code opens your project, it automatically reads CLAUDE.md from the project root. The file provides:

• Domain detection — Claude detects Connect usage from package.json, import patterns, and API call signatures
• Routing rules — deterministic IF/THEN rules that select llms-connect-full.txt and never route to UI framework docs
• API mapping — translates user intents to the correct Connect REST endpoints and OAuth flows
• Anti-patterns — explicit WRONG → RIGHT examples that prevent mixing Connect API calls with UI component code

Step 2: Configure the MCP server (Optional)

For live endpoint and SDK lookups, add the Mobiscroll MCP server:

claude mcp add --transport http mobiscroll https://mcp.dev.mobiscroll.com/

VS Code extension

If you are using the Claude Visual Studio Code extension, the server will not appear unless it is added with project scope. See the next command below.

To share the server with your team automatically, use project scope:

claude mcp add --transport http --scope project mobiscroll https://mcp.dev.mobiscroll.com/

This creates or updates .mcp.json in your project root. You can also create that file manually:

{
  "mcpServers": {
    "mobiscroll": {
      "type": "http",
      "url": "https://mcp.dev.mobiscroll.com/"
    }
  }
}

Scope	CLI flag	Config location	Shared with team
local (default)	--scope local	~/.claude.json	No
project	--scope project	.mcp.json in project root	Yes, via version control
user	--scope user	~/.claude.json	No, all your projects

info

Use --scope project for team repos so everyone gets the MCP server automatically when they clone the project.

Verify the connection: Run /mcp inside Claude Code. The panel lists each connected server and its tool count. A healthy connection shows mobiscroll with its tools, including the Connect-prefixed ones.

How it works

With CLAUDE.md in place and the MCP server configured, when you ask Claude Code to write Connect code it:

1. Detects your SDK language and version via resolveConnectEnvironment
2. Looks up the endpoint schema (getConnectEndpointSchema) or SDK method (getConnectSdkMethod) before writing any call
3. Uses mapConnectEndpointToSdk when translating a REST endpoint into SDK code, and getConnectErrorTaxonomy when writing error handling

So Claude always uses the current Connect API and SDK signatures, never hallucinated or outdated ones.

Domain isolation

Critical

Mobiscroll Connect is a backend integration layer — OAuth 2.0, REST API, webhooks, backend calendar sync. It has no UI components. Never mix Connect docs with UI framework docs.

Why this matters:

• Connect uses mobiscroll-connect.mdc and llms-connect-full.txt — never the UI framework files
• Mixing Connect docs with UI docs causes the AI to conflate REST endpoints with component APIs
• Connect has no JSX, no frontend framework bindings, no CSS

Rules:

1. Add only one .mdc file per project — the one matching your framework or domain
2. Register only one documentation source in Cursor
3. The MCP server is unified — the single mobiscroll server serves both UI and Connect tools. Isolation happens at the tool level: the Connect tools are all prefixed Connect (listConnectEndpoints, getConnectSdkMethod, …), so they never collide with the UI component tools. The .mdc and CLAUDE.md rules keep the AI on the Connect tools for Connect work.
4. If your project uses both a UI framework and Mobiscroll Connect, use separate AI context directories for each
5. If an AI assistant generates UI component code when you asked about Connect, check that mobiscroll-connect.mdc is active

Example queries

These examples show the kind of questions the AI integration is designed to handle correctly.

How do I authenticate a user with the Mobiscroll Connect OAuth flow?
How do I list all calendars for a connected Google account?
How do I create an event in an Outlook calendar via the Connect API?
What's the Python SDK call equivalent to POST /event?
How do I subscribe to webhook notifications for calendar changes?
What scopes are required for read-write calendar access?

Troubleshooting

AI generates UI component code instead of Connect API calls

Symptom: You asked about backend calendar sync or OAuth but the AI generates JSX components like <Eventcalendar /> instead of Connect REST API calls.

Fix: Verify that mobiscroll-connect.mdc is in place and that the registered @docs source in Cursor points to llms-connect-full.txt — not a UI framework file. The mobiscroll MCP server exposes both UI and Connect tools, so make sure the AI is using the Connect-prefixed tools (e.g. getConnectEndpointSchema) rather than the UI component tools. Connect and Eventcalendar are entirely separate products.

AI invents non-existent endpoints or parameters

Symptom: The AI suggests REST endpoints, request parameters, response fields, or SDK methods that don't exist in the Connect API.

Fix: The .mdc rules instruct the AI to only use APIs found in the Connect docs. If this still happens, explicitly reference @docs in Cursor queries, or verify that CLAUDE.md is in the project root for Claude Code. For the highest accuracy, enable the Mobiscroll MCP server so the AI fetches live endpoint and SDK schemas via the Connect tools instead of guessing. You can also ask the AI to confirm an endpoint exists in the Mobiscroll Connect docs.

MCP server does not appear after setup

Symptom: The MCP server panel shows no mobiscroll entry, or its Connect tools are not available.

Fix: Check that the config file is in the correct location and uses the correct root key — mcpServers for Claude Code and Cursor, servers for VS Code. Validate that the file is well-formed JSON, and that the server name is mobiscroll. For Claude Code, run /mcp to inspect connected servers; for the VS Code extension, add the server with project scope. If the server connects but the Connect-prefixed tools are missing, your deployed server may predate the Connect merge — reconnect after it is updated.

AI mixes Mobiscroll Connect with UI components

Symptom: The AI generates REST API calls when you asked about a frontend calendar component, or generates JSX/component code when you asked about the Connect API.

Fix: Mobiscroll Connect is a backend integration layer (OAuth, REST, webhooks) and has no UI components. Eventcalendar is a frontend UI component with no REST API. They use entirely separate .mdc files, documentation sources, and MCP servers. Verify that the correct .mdc file is active for your project. If you use both in the same codebase, keep separate AI context directories for each.

File reference

All AI integration files and endpoints are available at the following URLs:

Documentation files

File	URL
Connect	https://dev.mobiscroll.com/docs/llms-connect.txt
Connect (full)	https://dev.mobiscroll.com/docs/llms-connect-full.txt

Rules files

File	URL
Connect rules	mobiscroll-connect.mdc

Routing file

File	URL
Claude Code context	connect/CLAUDE.md

MCP server

Server	URL
Mobiscroll MCP (UI + Connect)	https://mcp.dev.mobiscroll.com/

File contents

The complete contents of each file are shown below. You can copy directly from these blocks or use the download links above.

CLAUDE.md

View CLAUDE.md

Loading...

Rules files (.mdc)

View mobiscroll-connect.mdc

Loading...