Managing 10+ TikTok Shop Clients Without Spreadsheets: The Claude Projects Architecture

High-growth digital agencies and performance marketing teams face a scaling ceiling that has nothing to do with creative talent or client demand. Instead, the bottleneck is purely operational. When an agency scales from managing two or three TikTok Shop clients to ten or more, the legacy workflow of spreadsheets, scattered CSV files, and manual creator tracking collapses under its own weight.

Each client represents a highly specific, non-negotiable set of constraints: distinct product Stock Keeping Units (SKUs), rigid brand guidelines, custom creative angles, and isolated creator rosters. When a single agency team manages these accounts across a shared database or a collection of shared spreadsheets, context collision is inevitable. Marketers accidentally pitch creators using the wrong brand voice, roster cross-contamination occurs when the same creator is pitched for conflicting campaigns, and spreadsheet sprawl turns daily tracking into a slow, error-prone manual chore.

To break through this operational ceiling, modern digital-native agencies are adopting a new, protocol-driven architecture. By combining Claude Projects, which act as isolated project knowledge bases, with a unified tool-calling model using the Model Context Protocol (MCP), agencies can build a secure, segregated, and highly automated multi-client engine. This guide details how to implement the Claude Projects architecture to manage ten or more active TikTok Shop clients with zero spreadsheets, absolute context isolation, and protocol-level security.

1. The Multi-Client Scaling Trap (Context Collision and Spreadsheet Sprawl)

For performance marketing agencies, the transition from boutique operations to enterprise scale is often a painful process marked by decreasing margins and increasing administrative overhead. In a traditional agency setup, creator sourcing relies on three highly fragmented layers:

• Shared Enterprise Databases: Agencies pay steep monthly fees to access centralized directories of pre-cached creator profiles. These databases treat all industries and campaigns as a single, generic pool, offering no native way to segregate client campaigns or manage client-specific constraints.
• Manual Spreadsheet Tracking: To customize recommendations, regional campaign managers export large CSV files of creator profiles, manually scrubbing columns to align with specific client briefs. As campaigns progress, these spreadsheets must be constantly updated with live metrics, outreach statuses, and shipping details, creating a web of conflicting versions.
• Fragmented Outreach Sequences: Once a list is vetted, operators must copy-paste contact details into separate email tools, manually customizing templates for each creator. This manual chain of events introduces errors, makes it difficult to track global communication history, and increases the risk of double-pitching the same local creator for two different clients.

When managing ten or more clients, this manual model becomes unsustainable. The labor cost of maintaining spreadsheets increases exponentially with each new client, while the risk of brand-damaging errors rises. A single mistake, such as sending a pitch for a luxury travel brand to a creator using the template of a high-protein food supplement, can destroy brand credibility and lose a client overnight.

2. The Solution: The Claude Projects Architecture

The Claude Projects feature, native to Claude Professional and Team plans, provides a powerful solution to this operational bottleneck. By acting as an isolated knowledge sandbox, a Claude Project allows agencies to define a dedicated workspace for each client. Each project contains its own set of custom instructions, brand guidelines, product briefs, and uploaded files, ensuring that Claude has access to the exact context it needs for a specific client without any risk of cross-contamination.

Instead of navigating a massive, shared database or a series of unlinked spreadsheets, an agency operator simply switches to the relevant client project inside Claude. When the operator asks Claude to identify potential creators, the AI parses the specific client documents, understands the target audience, and builds a tailored query.

However, an isolated knowledge base is only as good as its access to real-world data. If Claude only relies on static files uploaded to the project, the operator is still forced to manually search for creators and paste their profiles back into the chat. To solve this, agencies are using the Model Context Protocol (MCP) to connect their Claude Projects directly to live social data.

By deploying @lobby/mcp-server, agencies can equip every Claude Project with real-time, structured access to social commerce intelligence. This integration allows Claude to act as an active operator, translating isolated client briefs into real-time queries, qualifying creators against live performance metrics, and preparing tailored campaigns, all within a single, secure environment.

3. Client Brief Isolation (Preventing Roster Cross-Contamination)

The core benefit of the Claude Projects architecture is the absolute segregation of client context. In a shared database, there is no technical boundary preventing a marketer from proposing a creator who is already under an exclusive contract with a competitor. In the Claude Projects model, each workspace is a clean slate.

To establish this isolation, agencies upload three primary assets to each client's Claude Project:

1. The Brand Identity Brief: This document outlines the brand's tone of voice, visual identity standards, target demographics, and a list of strictly banned content elements (such as political commentary or lower-quality video formats).
2. The Product SKU Ledger: A detailed index of the specific products active in the current TikTok Shop campaign. This includes product features, key selling points, retail pricing, and the specific creative angles assigned to each SKU.
3. The Active Roster History: A list of creators who have previously collaborated with the client, are currently in negotiations, or have been placed on a client-specific blocklist.

Because these files are stored in the project's private knowledge base, Claude maintains strict boundaries. When an operator switches to the project for a high-end travel brand, Claude's reasoning loop is entirely focused on luxury aesthetics, premium accommodations, and high-income audiences. If the operator then switches to a project for a direct-to-consumer health brand, Claude instantly drops the travel context and focuses on functional benefits, clean ingredient lists, and fitness demographics.

When executing creator discovery, Claude uses these isolated briefs to formulate search parameters. It translates the creative angles defined in the SKU ledger into precise, contextual query strings. By passing these tailored queries to the Lobby Model Context Protocol API, Claude retrieves a highly targeted list of creators who are actively discussing topics that match the client's exact niche.

4. Grounding the Technology: The On-Demand Retrieval Paradigm

A common misconception among agency teams is that AI discovery tools rely on massive, pre-scraped indexes of the entire internet. Many legacy SaaS platforms claim to continuously stream and index every video posted to social media, a claim that is technically impossible to sustain without massive data degradation, proxy rotation failures, and high subscription markup.

The Lobby architecture operates on a fundamentally different, grounded paradigm: On-Demand Retrieval.

Lobby does not claim to index or continuously stream social videos. Instead, it acts as a real-time, protocol-driven bridge between your reasoning engine and the live platforms. When Claude executes a tool call through the Lobby MCP server, the system performs a targeted, live search on-demand.

This on-demand retrieval ensures absolute freshness and accuracy:

• Zero Stale Profiles: Instead of showing cached engagement rates from three months ago, Lobby pulls live, active platform data at the exact moment of the query. If a creator has pivoted their content style or stopped posting, the system identifies this immediately.
• Semantic Context Match: Lobby retrieves the actual spoken video audio, on-screen text, and written captions of the top-performing videos matching the search query. Claude can then analyze this raw, structured context to verify if the creator's true voice aligns with the client's guidelines.
• Efficient Token Usage: Because Lobby returns clean, structured JSON payloads directly to the MCP client, it avoids the massive token wastage associated with feeding raw HTML or unformatted web scrapes into the context window. This protocol-driven efficiency is crucial when managing ten or more clients, as it keeps API costs low and prevents model hallucinations.

Once the on-demand search retrieves a list of potential candidates, the system initiates the critical verification layer: the 10-video evidence audit.

5. The 10-Video Evidence Audit: Verification Over Hope

In social commerce, hiring a creator based on a single viral video is a recipe for poor campaign return on investment. Sponsoring a creator who had one video hit a million views but routinely averages three hundred views on their other posts results in highly volatile, expensive customer acquisition costs.

To protect client budgets and ensure reliable performance, the Claude Projects architecture enforces a mandatory validation step: The 10-Video Evidence Audit.

When Claude retrieves potential creators via the Lobby MCP server, it does not evaluate them based on total follower count or a single performance metric. Instead, the model analyzes a detailed performance ledger representing the creator's last ten published videos.

This audit, which is embedded directly into the final client reports, evaluates three critical criteria:

1. View Stability and Distribution: Claude calculates the median view count across the last ten videos, identifying if the creator maintains consistent baseline views or relies entirely on occasional viral anomalies. A stable median indicates a highly engaged, loyal audience.
2. Spoken Context Consistency: The audit parses the spoken transcripts of the last ten videos to verify that the creator naturally and consistently discusses the client's product category. If a creator claims to be a fitness influencer but nine of their last ten videos are about gaming, the model flags the profile as misaligned.
3. Creative Style Alignment: Claude reviews the visual structure of the videos, assessing whether the creator uses professional camera equipment, natural lighting, and clear product demonstrations, or relies on low-quality, automated slides and generic voiceovers.

By automating this 10-video audit inside the Claude Project, agencies can generate detailed, evidence-backed reports for their clients. Instead of proposing a list of names and hoping they deliver, operators can present clients with a verified performance audit for every recommended creator, proving consistent delivery before a single dollar is spent.

6. Sourcing via Spreadsheets vs. Claude Projects & Lobby MCP

To understand the financial and operational impact of transitioning to a protocol-driven architecture, it is helpful to compare the unit economics of a traditional spreadsheet-based agency workflow against the Claude Projects and Lobby MCP model.

The table below outlines the operational benchmarks for a digital agency managing ten active TikTok Shop clients, with an average of twenty creator activations per client, per month.

(Note: Illustrative workflow model representing operational benchmarks across digital agency operations.)

By shifting the heavy lifting of data retrieval, formatting, and initial vetting to the protocol layer, agencies can reduce their manual sourcing overhead by over ninety percent. This drastic reduction in labor hours allows a single agency operator to easily manage ten or more active clients, shifting their focus from tedious administrative coordination to high-value creative strategy and client relationships.

7. Operationalizing the Workspace: A Practical Setup Blueprint

Setting up your multi-client agency engine requires configuring your local workspace and establishing clear, consistent instructions for your Claude Projects. Below is a practical step-by-step blueprint to initialize your architecture.

Step 1: Initialize the Unified @lobby/mcp-server

To give Claude Desktop or Cursor access to the Lobby discovery engine, you must configure your local Model Context Protocol settings. This single, unified server handles all queries across your workspace.

Open your local configuration file (e.g., claude_desktop_config.json or your Cursor settings) and add the following server configuration:

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

This configuration integrates the Lobby creator activation platform directly into your workspace. When any Claude Project triggers an MCP tool call, the request is securely routed through this local protocol bridge.

Step 2: Establish the Client Claude Project

Inside Claude, create a new Project for each client. For example, if you are managing a premium lifestyle brand called "Solace Travel," name the project Solace Travel - TikTok Shop.

Upload your client-specific documentation directly into the Project's Knowledge Base. This should include: * solace_travel_brand_guidelines.md (Tone, visual style, banned topics) * solace_travel_sku_catalog.json (Product features, hooks, and retail pricing) * solace_travel_roster_history.csv (Previous creators, active blocklist, exclusive agreements)

Step 3: Apply the System Instructions (The Agentic Sourcing Prompt)

In the custom instructions text area of your Claude Project, paste the following system prompt. This prompt establishes the precise behavioral boundaries, grounding instructions, and tool usage rules for the client workspace:

By establishing this clear, programmatic system prompt, you ensure that Claude operates as a disciplined, client-specific execution agent. It will automatically translate the brand's requirements into highly accurate searches, execute the rigorous validation steps, and deliver publication-ready rosters with zero manual oversight.

8. Scaling to 10+ Clients (The Operational Playbook)

Once your initial client project is configured and verified, scaling the architecture to manage ten or more clients is a simple process of replication. Because the @lobby/mcp-server operates as a unified protocol layer, you do not need to install or configure new software for each client. The single local server serves all projects seamlessly.

To maintain order and high efficiency across a large client portfolio, agencies should follow three key operational guidelines:

• Establish a Bi-Weekly SKU Review: TikTok Shop trends and product inventory move quickly. Make it a rule to update each project's product catalog and active campaign briefs every two weeks. This ensures that Claude is always sourcing creators for active, in-stock products.
• Enforce Global List Hygiene: While Claude Projects are isolated to prevent context leaks, you must ensure that your outreach list is cleaned of duplicates. Use Lobby's built-in campaign coordination tools to enforce an organizational cool-off period, preventing different client workspaces from pitching the same creator within a short time window.
• Automate Reporting Delivery: Use Claude to compile the final verified rosters into clean, publication-ready markdown files. You can copy these reports directly into your client communications, providing clients with immediate proof of our rigorous vetting and validation process.

By combining the structural isolation of Claude Projects with the real-time, on-demand power of the managing TikTok creators in Claude Desktop via MCP workflow, agencies can completely eliminate the manual spreadsheet grind. Sourcing, auditing, and preparing outreach becomes a unified, frictionless protocol loop, allowing your team to focus on what truly matters: driving high-converting creator partnerships and scaling your clients' social commerce revenue.

Frequently asked questions

How do Claude Projects keep client data separate from other campaigns?

Claude Projects act as isolated, private knowledge sandboxes. Files, brand guidelines, and custom instructions uploaded to a specific client project are completely invisible to other projects. When an operator switches projects, Claude drops the active context and loads the new client's guidelines, ensuring zero risk of context leaks or roster cross-contamination across your client portfolio.

How does Lobby retrieve live TikTok data if it does not use a cached database?

Lobby operates on an on-demand retrieval model. Rather than relying on a static, pre-scraped database of profiles that quickly goes out of date, the Lobby engine queries live platforms in real-time when a tool call is executed. This ensures that the creator metrics, contact information, and posting history you receive are completely fresh and accurate at the exact moment of your search.

Why is a 10-video audit better than checking follower count or total views?

Follower counts and lifetime views are easily manipulated and often reflect historical performance rather than current engagement. A 10-video performance audit calculates the median view count, posting frequency, and content consistency across a creator's last ten published posts. This analysis provides a realistic picture of the creator's current audience reach and creative alignment, helping agencies protect client budgets from low-performing partnerships.

Can we connect our existing outreach tools directly to this Claude architecture?

Yes. Once Claude has curated and approved your verified creator roster, the profiles can be activated directly. By using Lobby's unified system, teams can draft personalized pitches and coordinate outreach through the direct inbox engine. This eliminates the need to export CSV lists, format data, and upload files into separate email platforms.

Does this architecture require technical development skills to maintain?

No. The beauty of the Model Context Protocol is its simplicity. The configuration is a straightforward JSON file, and the entire system is operated using natural language commands inside your daily reasoning workspace. Your existing agency operations team can easily manage the day-to-day workflow, update client files, and curate creator lists with zero programming knowledge.