Technical Walkthrough: Setting up @lobby/mcp-server in Cursor and Claude Desktop for TikTok Commerce

Digital-native performance agencies, direct-to-consumer operators, and modern growth teams are experiencing an architectural shift. They are moving away from fragmented, browser-based marketing dashboards and consolidating their entire growth loops within integrated development and reasoning environments like Cursor and Claude Desktop. By treating these AI-native clients as a unified console, engineers and marketers can write code, analyze data, and build campaign assets without leaving their primary workspace.

Sourcing, qualifying, and activating social creators has remained isolated. Influencer marketing workflows have been trapped inside expensive legacy databases, closed enterprise software, and manual spreadsheets. To resolve this friction, software teams are adopting the Model Context Protocol. By integrating @lobby/mcp-server into their local configurations, development and growth teams can connect live creator discovery, ten-video consistency validation, and direct outreach pipelines directly to their Large Language Model client. This integration replaces a manual series of steps with a protocol-driven pipeline.

1. The Bottlenecks of Legacy SaaS Directories and Static Data

For years, agency sourcing workflows have relied on pre-compiled, cached directories. Platforms like Modash, Upfluence, and Grin require agencies to pay premium subscription fees to search historical profile records. While these directories served early social campaigns, they fail in high-velocity TikTok commerce environments where creator focus, performance, and availability change rapidly.

These legacy directory models present three severe limitations:

• Stale and Outdated Profile Data: Creator statistics in massive databases are updated infrequently, sometimes only once every few months. In the fast-moving short-form video space, a creator's engagement metrics, posting frequency, and niche can shift in a week. Surfacing inactive profiles wastes engineering and research time.
• The Contact Retrieval Gatekeeper Problem: Traditional directories crawl public web pages to find general management or agency emails. When teams send outreach to these addresses, they are met with slow response times and high fee demands, bypassing the direct, agile relationships required for social commerce.
• Workflow Fragmentation and Spreadsheet Friction: Operators must run searches in a database, select candidate profiles, export a CSV file, manually clean formatting errors, and upload the list into separate email sequencing software. This manual chain introduces error and prevents real-time, personalized outreach.

This manual workflow limits execution speed. As agencies scale campaigns, the overhead of managing static lists and correcting data errors grows unsustainably.

2. The Limits of DIY Web Scraping and Browser Automation

To avoid expensive SaaS platforms, some software teams attempt to build custom scraping systems. They deploy open-source libraries like OpenClaw or write custom scripts using Playwright and Puppeteer to scrape social platforms, trying to solve the problem of feeding raw DOM HTML into Claude context for creator discovery.

While a custom scraper seems like an easy solution, it quickly hits practical limitations:

• Advanced Anti-Bot Defense Mechanisms: Social platforms utilize sophisticated anti-scraping systems, including canvas fingerprinting, TLS JA4 verification, and behavioral analysis. Headless cloud browsers running on standard datacenter IP ranges get blocked or hit with CAPTCHAs almost instantly. Maintaining a residential proxy network and updating scraper code to bypass these blocks becomes a full-time engineering task.
• Session Maintenance and Cookie Rot: Social platforms require active, authenticated user sessions for deep content discovery. Custom automated scripts struggle with session rot, which invalidates browser cookies and breaks data pipelines mid-campaign.
• Token Waste and Context Inflation: Passing raw, unformatted web pages or noisy markup into an LLM context window consumes expensive tokens. A single headless browser scrape can consume thousands of tokens on irrelevant styling, navigation scripts, and tracking code, increasing API costs while causing model hallucinations.

AI agents do not require a fragile browser wrapper to read raw elements. Instead, they require structured, clean, and reliable data provided directly at the protocol level.

3. Resolving Friction with the Model Context Protocol

The Model Context Protocol, open-sourced by Anthropic, establishes an open standard for connecting LLMs to external data sources and local tools. Instead of building custom integrations for every tool, developers write an MCP server that exposes tools via a standard JSON-RPC protocol over standard input/output.

The @lobby/mcp-server applies this protocol directly to social commerce. By running the Lobby server, your AI agent gains native capabilities to search, inspect, and qualify creators directly within its reasoning window.

Lobby's architecture is built on structured, on-demand queries rather than static caching:

• Real-Time Platform Sourcing: Lobby does not claim to index or stream entire social video platforms. Instead, it acts as a real-time retrieval layer. When an agent requests creator recommendations for a specific brief, Lobby queries live platform data on-demand to surface active accounts. This ensures that you are managing TikTok creators inside Claude Desktop via MCP using live, accurate data.
• Ten-Video Consistency Validation: To ensure creators are currently active and reliable, the server retrieves and analyzes metrics across their last ten videos. The agent evaluates view stability, engagement rates, video topics, and visual style consistency to verify the creator's true audience alignment.
• Direct Activation Integration: Once a creator is qualified, the server retrieves verified contact details from the Lobby database. Marketers can then draft and execute direct pitches without leaving the command line or IDE.

This protocol-driven approach provides clean, structured JSON data directly to the LLM, eliminating web scrapers, proxy networks, and raw HTML parsing.

4. The @lobby/mcp-server Tool Reference

The @lobby/mcp-server exposes a collection of tools designed to handle creator discovery, data validation, contact retrieval, and outreach copywriting. These tools allow your AI agent to execute complex sourcing workflows.

The primary tools available on the server are:

A. search_creators

The search_creators tool queries the live social graph on-demand to locate active accounts matching your campaign brief, location, and performance targets.

• Parameters:
  • query (string, required): The descriptive search query or brand brief.
  • location (string, optional): The geographic target for local campaigns.
  • min_followers (number, optional): The minimum follower limit.
  • max_followers (number, optional): The maximum follower limit.
• Expected Output: A JSON array of creator profiles, including username, bio, follower count, category, and average engagement rate.

B. verify_creator_videos

The verify_creator_videos tool retrieves performance metrics for the creator's last ten videos. The AI agent uses this data to evaluate content consistency and audience engagement, screening out bot-boosted accounts.

• Parameters:
  • username (string, required): The username of the creator to verify.
  • count (number, optional): The number of recent videos to analyze (default is 10).
• Expected Output: A JSON array containing individual video metrics, including view count, comment count, share count, video transcript, publishing date, and detected visual style keywords.

C. get_direct_contacts

The get_direct_contacts tool retrieves verified email addresses and communication channels for qualified creators. This tool connects directly to the Lobby database, bypassing gatekeepers and general contact forms.

• Parameters:
  • username (string, required): The verified creator's username.
• Expected Output: A JSON object containing the direct email address, associated agency details, and verified messaging handles.

D. generate_activation_pitch

The generate_activation_pitch tool generates highly tailored, personalized pitches for 1:1 activation based on the creator's video style, recent performance, and brand brief.

• Parameters:
  • username (string, required): The target creator's username.
  • brand_brief (string, required): The campaign value proposition and requirements.
  • selected_videos (array of strings, optional): Specific video IDs or topics to reference.
• Expected Output: A polished, context-aware pitch draft that references specific elements of the creator's recent videos to establish authenticity.

5. Walkthrough: Configuring the Lobby MCP Server

To integrate the Lobby MCP server into your daily workspace, you must have Node.js version 18 or higher installed on your local machine. You will also need to generate your Lobby API key to authenticate with the platform.

A. Configuring Claude Desktop

To add the Lobby server to Claude Desktop, you must modify your local claude_desktop_config.json configuration file. This file is located at one of the following paths:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

Open the file in your preferred text editor and add the Lobby configuration inside the mcpServers object:

{
  "mcpServers": {
    "lobby-creator-activation": {
      "command": "npx",
      "args": [
        "-y",
        "@lobby/mcp-server"
      ],
      "env": {
        "LOBBY_API_KEY": "your_secure_lobby_api_token_here"
      }
    }
  }
}

Once saved, restart Claude Desktop. You will see a plug icon in the lower right corner of the chat input box, indicating that Claude has successfully loaded the four Lobby tools.

B. Configuring Cursor IDE

To configure the Lobby server globally for Cursor, navigate to Cursor Settings > Features > MCP. Click + Add New MCP Server and enter the following settings:

• Name: lobby-creator-activation
• Type: command
• Command: npx -y @lobby/mcp-server

Provide your API key in the environment variables block as LOBBY_API_KEY=your_secure_lobby_api_token_here.

For team-based workflows and per-project setups, you can configure Cursor to load the server automatically from your repository. Create a file named .cursor/mcp.json in your project's root directory:

{
  "mcpServers": {
    "lobby-mcp-server": {
      "command": "npx",
      "args": [
        "-y",
        "@lobby/mcp-server"
      ],
      "env": {
        "LOBBY_API_KEY": "your_secure_lobby_api_token_here"
      }
    }
  }
}

This project-scoped configuration allows any team member opening the repository in Cursor to immediately access the Lobby tools, ensuring a consistent setup across your entire team.

6. Sourcing and Activating Creators with Agentic Prompts

With @lobby/mcp-server configured, your AI agent can manage the entire influencer sourcing and activation lifecycle through natural-language prompts. Here is an example of an interactive session between a growth operator and an integrated AI agent.

Phase 1: Context-Driven Sourcing

User Prompt:

"I need to find five active micro-creators in New York City for our gourmet kitchenware brand. They should post content about home baking, sourdough tutorials, or kitchen styling. Their follower count must be between 15k and 100k, and they must have high video view consistency."

Agent Execution: The agent recognizes the request and calls the search_creators tool:

{
  "query": "home baking sourdough tutorials kitchen styling",
  "location": "New York, NY",
  "min_followers": 15000,
  "max_followers": 100000
}

The tool returns a list of matching creators. The agent processes this data and presents the top candidates:

"I have retrieved five matching sourdough and home-baking creators in New York City from the live social graph. Here are the top three candidates:

1. bakes_with_clara (42.5k followers) - Specializes in artisan sourdough and retro kitchen design.
2. crust_and_crumb_nyc (28.2k followers) - Focused on sourdough science and recipe development.
3. styling_my_stove (18.9k followers) - Focuses on aesthetic stove-top setups and minimal kitchen organization."

Phase 2: Ten-Video Performance Verification

User Prompt:

"Verify the view stability and engagement consistency for 'bakes_with_clara' over her last ten videos. Check if she has regular posting habits and confirm that her content is naturally aligned with kitchenware."

Agent Execution: The agent calls the verify_creator_videos tool to inspect her latest video metrics:

{
  "username": "bakes_with_clara",
  "count": 10
}

The tool returns the performance data for her last ten videos. The agent performs an analysis:

"I have analyzed the last ten videos published by bakes_with_clara. Here is the verification report:

• View Stability Analysis: Her average view count is 14.2k, with a standard deviation of 2.1k. This indicates consistent view stability, with no single viral video distorting her average.
• Posting Frequency: Her last ten videos span 18 days, showing an active posting habit of 3 to 4 videos per week.
• Content Alignment: Out of the ten videos, eight feature aesthetic baking tutorials. She regularly features high-end mixers, proofing baskets, and baking steels, confirming that her content is naturally aligned with kitchenware.
• Audience Engagement: Her comment-to-view ratio is 4.8 percent, which is significantly higher than the industry average of 1.5 percent for this follower range. The comments focus on recipe details and baking equipment."

Phase 3: Direct Activation & Personalized Outreach

User Prompt:

"Get the direct contact email for 'bakes_with_clara'. Then, write a personalized outreach email introducing our brand. Reference her recent sourdough proofing video and emphasize that we want to send her our new professional baking stone."

Agent Execution: The agent first retrieves the verified email using get_direct_contacts:

{
  "username": "bakes_with_clara"
}

The tool returns clara.bakes.nyc@example.com. The agent then calls generate_activation_pitch to craft the email draft:

{
  "username": "bakes_with_clara",
  "brand_brief": "Premium professional baking stone made from cordierite clay, launching next month. We are offering a free unit for review with no strict posting requirements.",
  "selected_videos": ["artisan_sourdough_proofing_v12"]
}

The tool generates a tailored, context-aware pitch draft. The agent presents the final copy to the user:

Subject: NYC sourdough routine x cordierite baking stone To: clara.bakes.nyc@example.com

Loved your rise technique with the retro cane basket. We build professional cordierite baking stones that retain heat for artisan crusts. Can we send one to your NYC kitchen + $250 flat for an honest review?

Because the agent reads the actual performance and content details of the creator's recent videos, it avoids generic templates that creators typically ignore.

7. Sourcing Workflows: Operational Benchmarks

By moving from legacy browser-based directories and manual spreadsheets to an AI-native, protocol-driven workflow, agencies achieve significant efficiency gains. The table below compares the two approaches:

(Note: Illustrative workflow comparison model representing typical performance improvements across digital agency operations.)

Implementing this protocol-driven workflow reduces the time required to source, verify, and pitch a creator from hours to minutes. This speed allows marketing teams to focus on strategy, creative direction, and campaign relationships.

Frequently asked questions

Does the @lobby/mcp-server index or store video files?

No. Lobby is built as a custom-intent search and activation layer. It does not index, cache, or stream entire social video files. Instead, it processes your search query in real time, requests live data from the platforms, and extracts key creator metadata and performance metrics on-demand to maintain high speed and data freshness. Complete developer specs can be reviewed in the official Lobby API and MCP documentation.

Can I run multiple brand campaigns with a single Lobby MCP integration?

Yes. By utilizing Cursor Workspace configurations or Claude Projects, you can set up separate workspaces for each brand. Each project maintains its own instructions, pitch templates, and list of qualified creators, allowing a single LLM client to manage multiple campaigns without data pollution.

How does the server verify a creator's authenticity?

The @lobby/mcp-server retrieves performance metrics from the creator's last ten videos. The agent then analyzes view consistency, comments, and engagement metrics to flag abnormal patterns, such as sudden spikes in views without matching comments, or sudden shifts in content topics. This automated check helps you avoid wasting budget on inactive or bot-boosted accounts.

Why is an MCP-based workflow better than exporting CSV lists?

Exporting CSV lists from a static database introduces data rot, requires manual formatting, and separates discovery from your outreach tools. An MCP-based workflow keeps your data live and integrated. The agent can search, verify, and draft custom pitches in one continuous flow, removing spreadsheets from the process and keeping your campaign data organized.

How do I get started with the Lobby MCP server?

You need an active Lobby account and an API key. Once you have your key, configure the server in Cursor or Claude Desktop using the standard @lobby/mcp-server package. For a step-by-step walkthrough, see our tutorial on installing @lobby/mcp-server in Cursor and refer to the official Lobby API and MCP documentation for API details.