yandex-direct-mcp
A Yandex Direct MCP server with about 80 tools, from reading stats to budgets, bids and launching a campaign, guarded by project and allow_write flags
High risk
We rate an entry high when the tool writes to external systems, handles money, production databases or secrets, or runs arbitrary commands. The CLI installs it only with your consent.
Why this level
- Write tools change budget, bids, campaign status and ad text
- The direct_rpc escape tool calls any API method with no extra restrictions
- One OAuth token grants access to the entire ad account across all brands
Install
Manual install
git clone https://github.com/gmen1057/yandex-direct-mcp.git
cd yandex-direct-mcp
npm install
cp .env.example .env
npm run smokeInstall from source and run the smoke test before first use.
This is third-party code. Review the repository files before installing.
What it does
The server exposes about 80 tools for the Yandex Direct API v5: reading campaigns, ad groups, keywords, ads, negative keywords and excluded sites, performance reports with spend, clicks and CTR broken down by dimension and search queries, and write operations covering budget, pause and resume, bids, autotargeting, the max-conv strategy, ad text and images, and a batch launch pack from campaign to moderation. An escape tool, direct_rpc, calls any API method directly. Every write call requires an explicit allow_write flag and a project binding from projects.json, so a shared token cannot accidentally touch another brand, and defaults to a dry-run plan.
Who it is for. For agencies and PPC specialists running several brands on one Yandex Direct token.
Good fit when
- You need a quick report on spend, clicks and search queries without opening the Direct interface
- You want to manage budget, bids and campaign status from a conversation with an agent
- You run several projects on one OAuth token and need protection against editing the wrong brand by mistake
Not a fit when
- You need read-only access with zero write risk: this server mixes read and write under one token
- You have no time to set up projects.json and understand the allow_write flag before the first edit
Example request
Show me the weekly campaign report for the shop project and suggest negative keywords to addLimitations
One OAuth token opens the entire ad account; brand isolation relies on correctly filling projects.json, not on API permissions. There is no npm package, install is from source. The author states the project is not affiliated with Yandex. API amounts are in micro-units while the tools work in rubles, and combining DailyBudget with WeeklySpendLimit in one call returns error 4004.
How to disable. Remove the server from your MCP client config and revoke the Yandex Direct OAuth token at oauth.yandex.ru.
MCP
- Transport
- stdio
- Authentication
- API key
| Environment variables | |
|---|---|
| YANDEX_DIRECT_TOKEN required, secret | OAuth token for an application with Yandex Direct API rights |
| YANDEX_DIRECT_CLIENT_LOGIN | Client login for agency access |
Security check
- Write tools change budget, bids, campaign status and ad text
- The direct_rpc escape tool calls any API method with no extra restrictions
- One OAuth token grants access to the entire ad account across all brands
README in short
The Russian README describes the server as built in production by the Arckep AI studio, with a quick start via npm install and npm run smoke, config examples for Claude Desktop, Claude Code and Cursor in the examples folder, and an AGENTS.md guide for an agent to install it unattended. It separately covers write-safety rules, the projects.json format for brand tagging, and Direct API quirks: micro-units, the 4004 limit error, headline moderation words, and int64 precision for ad ids. MIT license; the author notes no affiliation with Yandex.
FAQ
How does it prevent accidentally editing the wrong brand?
Write calls require a project key from projects.json, all is not allowed, and without the allow_write flag only a dry-run plan runs.
Can it call an API method with no dedicated tool?
Yes, the direct_rpc escape tool calls any service.method of the v5 or v501 API directly.
Related
Official DataLens (Yandex) skills for Claude Code, Codex and OpenCode: SDK, HTML reports and RLS resolution
A Yandex Metrika MCP server generated from the API spec: 108 methods, 10 tools declared by default, transparent filters and response metadata
An open MCP server and agent skill set for SEO: keyword research, competitors, backlinks and site audits powered by DataForSEO
A Claude Code plugin for SEO audits: technical SEO, E-E-A-T, schema, local and AI search via parallel subagents