LinkedIn Prospecting CSV Manager (MCP Server)

A high-performance, token-efficient Model Context Protocol (MCP) server designed for managing LinkedIn prospecting data in local CSV files. Built with Python and pandas.

Why This Project?

Managing large CSV files directly within an LLM (like Claude or ChatGPT) is inefficient and error-prone:

• Token Drain: Reading a 1000-line CSV can consume ~30,000 tokens per operation.

• Data Corruption: Manual file writing by LLMs often leads to escaping issues or column mismatches.

• Scalability: LLMs struggle with O(n) deduplication and full-file rewrites.

This MCP server reduces token usage by 60x+ by offloading CSV logic to your local machine.

Related MCP server: MCP VectorStore Server

🛠️ Features (V2)

• Standardized Golden Schema: Enforces a consistent set of columns across all your prospecting campaigns.

• "Auto-Repair" Header Normalization: Automatically renames legacy or inconsistent headers (e.g., v2 Score -> match_score) to match the Golden Schema.

• Atomic Writes: Uses temporary-file-and-replace patterns to ensure zero data corruption during file updates.

• Efficient Appending: Add new profiles with automatic deduplication based on linkedin_url.

• Multi-Value Filtering: Query profiles by Score, Company, or multiple Locations (e.g., ["USA", "Canada"]).

• Full-Text Search: Case-insensitive search across all text fields.

• Absolute Path Enforcement: Prevents "ghost files" by resolving all paths reliably.

The Golden Schema

Every CSV processed by the server is automatically standardized to:

1. full_name

2. linkedin_url (Deduplication Primary Key)

3. headline

4. company

5. company_size

6. location

7. match_score

8. match_reason

9. current_role_mention

10. found_date

11. icp_source

📦 Installation

Prerequisites

• Python 3.10+

• uv (recommended for dependency management)

Local Setup

1. Clone this repository:

   git clone https://github.com/denis911/antigravity-mcp-csv-add-deduplicate.git
   cd antigravity-mcp-csv-add-deduplicate

2. Install dependencies:

   uv sync

Claude Desktop Integration

Add the following to your Claude Desktop configuration file:

• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "linkedin-prospecting-csv": {
      "command": "uv",
      "args": [
        "--directory",
        "C:\\path\\to\\your\\repo\\antigravity-mcp-csv-add-deduplicate",
        "run",
        "python",
        "-m",
        "linkedin_prospecting_csv.server"
      ]
    }
  }
}

🧪 Testing

Automated Testing

We use pytest with real-world data from the TESTS directory:

uv run pytest TESTS/test_csv_ops.py

🛠️ Available Tools

🔒 Security & Privacy

This server runs locally on your PC. Your CSV data never leaves your environment; only the specific results of your queries (filtered rows or stats) are sent to the LLM.