Build on it

MCP server

Connect Claude Code, Claude Desktop, Cursor or VS Code to your DL2App links through the hosted Model Context Protocol server.

DL2App hosts a Model Context Protocol server at /api/mcp on the dashboard’s host. It lets a coding agent or assistant list, create, change and archive links and read analytics, with the same keys, scopes, limits and errors as the REST API.

Connect

Authentication is a secret key in a header: Authorization: Bearer sk_live_…. Create the key under API keys in a project. Its scopes decide what the assistant can do: a key without links:write can only read.

Claude Code

Run in a terminal. Add --scope user to use it in every project.

claude mcp add --transport http dl2 https://portal.dl2.app/api/mcp \
  --header "Authorization: Bearer sk_live_…"

Claude Desktop

Settings → Developer → Edit Config (claude_desktop_config.json), then restart Claude Desktop.

{
  "mcpServers": {
    "dl2": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://portal.dl2.app/api/mcp",
        "--header",
        "Authorization:${DL2_AUTH}"
      ],
      "env": {
        "DL2_AUTH": "Bearer sk_live_…"
      }
    }
  }
}

Claude Desktop's built-in connectors sign in with OAuth, which DL2App doesn't have yet, so this uses the mcp-remote bridge (needs Node.js) to send the key as a header.

Cursor

.cursor/mcp.json in your project, or ~/.cursor/mcp.json for every project.

{
  "mcpServers": {
    "dl2": {
      "url": "https://portal.dl2.app/api/mcp",
      "headers": {
        "Authorization": "Bearer sk_live_…"
      }
    }
  }
}

VS Code

.vscode/mcp.json in your workspace. VS Code asks for the key once and stores it.

{
  "servers": {
    "dl2": {
      "type": "http",
      "url": "https://portal.dl2.app/api/mcp",
      "headers": {
        "Authorization": "Bearer ${input:dl2-key}"
      }
    }
  },
  "inputs": [
    {
      "type": "promptString",
      "id": "dl2-key",
      "description": "DL2App secret key (sk_live_…)",
      "password": true
    }
  ]
}

The command-line tool prints the same configuration with your key filled in: dl2 mcp config --client cursor.

Transport

Streamable HTTP, request and response only: POST /api/mcp with one JSON-RPC 2.0 message, or a batch, returns one JSON answer. There is no event stream (GET answers 405) and no session, so any request can stand alone. Protocol versions: 2025-11-25, 2025-06-18, 2025-03-26, 2024-11-05.

Tools

There is one tool for each REST endpoint, generated from the same list as the API reference, plus a few shortcuts for assistants.

Tool	Scope	Kind	What it does
`list_projects`	`links:read`	read	Projects the key can see, newest first.
`get_project`	`links:read`	read	One project with its link hosts, a summary of its iOS and Android apps, and how many links it has.
`list_links`	`links:read`	read	A project's links, newest first.
`create_link`	`links:write`	write	Creates one link.
`create_links_bulk`	`links:write`	write	Creates up to 1,000 links in one request.
`get_link`	`links:read`	read	One link by id.
`update_link`	`links:write`	write, destructive	Changes only the fields you send; the rest stay as they are.
`archive_link`	`links:write`	write, destructive	Archives the link: it stops redirecting and disappears from lists, and its click history is kept.
`get_link_stats`	`analytics:read`	read	Totals, a daily series and breakdowns for one link.
`get_project_stats`	`analytics:read`	read	The same figures across the whole project, plus its ten most-clicked links and the clicks on paths that aren't links (passthrough).
`get_project_health`	`links:read`	read	The most recent result of each check on each of the project's hosts.
`create_campaign_links`	`links:write`	write	Creates one link per channel for a campaign: the same destination each time, a different set of UTM tags.
`find_links`	`links:read`	read	Searches a project's links with everyday filters; every filter given must match.
`explain_link`	`links:read`	read	Says where a link sends each kind of visitor (iPhone, iPad, Android, in-app browsers, desktop, link previews), after applying the project's defaults and which apps are set up.
`qr_code`	`links:read`	read	Returns the QR code for a link as SVG, drawn with the design saved for it in the dashboard (or the default look).

Safety

Dry runs. Every write tool takes dry_run. With it, nothing is written and the result says what would change: the links that would be created, a field-by-field diff for an update, or the link and its recent clicks for an archive. create_links_bulk, archive_link, create_campaign_links are dry runs unless you send dry_run: false.
Scopes. Write tools need links:write, dry run or not. Stats tools need analytics:read.
Annotations. Tools carry readOnlyHint, destructiveHint and idempotentHint, so a client can ask before the destructive ones.
Retries. Write tools take idempotency_key; a repeat within 24 hours returns the first result.
Small results. Lists return compact rows, 25 at a time, with next_cursor.
Errors. A failed call has isError: true and the API’s error code, a message and a suggested next step.

Resources and prompts

dl2://projects: the projects the key can use.
dl2://projects/{project_id}/settings: one project’s hosts, apps, default routing, deferred deep linking and link-preview defaults.
Prompt launch_campaign: Plans and creates a set of tracked links for one campaign, with a dry run first.
Prompt audit_broken_links: Checks a project's host health and looks for links that are disabled, expired, flagged or pointing nowhere useful.
Prompt weekly_link_report: Summarises the last seven days of clicks, app opens and installs for a project.