SergeyKrin9/mcp-servers

🐍 Python · 🇷🇺 Russian market · 🤖 8 server directories

A collection of 8 custom Model Context Protocol (MCP) server directories that wire marketing platforms into LLM-agent loops. Built so an agent (Claude Code, Claude Desktop, Cursor, Windsurf) can run ad campaigns, read analytics, publish posts, and check SEO health directly — without humans clicking buttons.

If you don't know what MCP is, start here — it's Anthropic's open protocol for plugging tools into LLM apps.

What's inside

Together they cover the full growth stack for a typical Russian-market product: paid ads (Direct + Telegram Ads + VK Ads via Click.ru), organic search (Yandex Webmaster + GSC), and content distribution (VK + Telegram).

Related MCP server: datavessel

Why this matters

In 2026 the bottleneck of growth isn't "what to do" — agents can plan. It's the glue between AI plans and platform APIs. These servers are that glue.

Start by listing the selected server’s tools. Review their parameters before using real accounts: several tools create campaigns or publish content.

⚡ Quickstart (3 steps)

1. Clone & install:

git clone https://github.com/SergeyKrin9/mcp-servers.git
cd mcp-servers/yadirect            # or any other server
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt    # if present
# yadirect reads YANDEX_DIRECT_TOKEN from the process environment.
# Configure it in your host; no .env.example or dotenv loader is provided here.

2. Wire into your AI host:

{
  "mcpServers": {
    "yandex-direct": {
      "command": "/abs/path/to/mcp-servers/yadirect/.venv/bin/python3",
      "args": ["/abs/path/to/mcp-servers/yadirect/server.py"],
      "env": { "YANDEX_DIRECT_TOKEN": "your-token" }
    }
  }
}

{
  "mcpServers": {
    "vk": {
      "command": "/abs/path/to/.venv/bin/python3",
      "args": ["/abs/path/to/mcp-servers/vk/server.py"],
      "env": { "VK_TOKEN": "your-token", "VK_GROUP_ID": "123456" }
    }
  }
}

Identical config — Windsurf uses the same MCP server format.

3. Check tool discovery. Ask your host to list the configured server’s tools without invoking them. This checks MCP registration without making advertising or publishing requests. API integration checks require a separate test account and are not covered by this step.

Credentials needed

yandex-webmaster/oauth_helper.py includes a working OAuth flow if you want to grab tokens locally.

GSC configuration migration

The GSC server no longer falls back to a machine-specific credential path. Set GSC_SERVICE_ACCOUNT_KEY in the environment of the MCP server process to the absolute path of your existing service-account JSON file:

{
  "env": {
    "GSC_SERVICE_ACCOUNT_KEY": "/absolute/path/to/your-existing-service-account.json"
  }
}

Merge this env entry into your existing GSC server configuration and restart that server. Replace the placeholder with your own local path; do not paste the JSON file contents into the configuration or commit them. This change creates no credentials and grants no new permissions. The existing service account must already have access to the Search Console property. If the variable is missing, the server reports a configuration error when a GSC tool is called.

Tech stack

• Python 3.10+ · stdlib + requests

• mcp Python SDK

• One process per platform — easy to enable/disable independently

Roadmap

• Telegram Ads MCP (independent of Click.ru, native API)

• Plausible / Yandex.Metrika real-time MCP (currently only reports)

• One-shot CLI wrappers so you can call any of these without an MCP host

⭐ Star history

📚 Related work

• agnis-workflow — the methodology that uses these MCP servers as first-class tools

• broker-reels — Python pipeline for real estate videos, with an offline demo

• awesome-mcp-servers — the canonical MCP catalog

Author: Sergei Grin · sergeykrin98@gmail.com · GitHub

If you ship to a Russian audience and use Claude — drop a ⭐. It tells me to keep adding servers.