Access levels of skills and MCP servers

The access level of every catalog entry and the reasons behind it. The level shows what a tool adds to what the agent can already do. It does not rate code quality.

Broad access

    • Places real orders and charges the saved card
    • Stores Yandex account session cookies
    • Uses a private API, which may break Yandex terms
    • Following the skill, the agent can change prices, stocks and order statuses in the live store
    • The API has no sandbox
    • It requires the seller cabinet Api-Key
    • Changes prices, listings and orders in a live seller account
    • Some methods are irreversible, such as deleting products
    • The key grants access to the whole account
    • By default it can create, change and stop Direct campaigns and bids
    • It can change Metrika goals and counters and delete Webmaster sitemaps
    • Start with YD_READONLY=true and enable writes only deliberately
    • Gets OAuth access to account mail, disk, calendar and tasks
    • Can send emails and publish files
    • Runs Python scripts that call Yandex APIs
    • In Direct live mode it creates campaigns and changes budgets and bids, spending real money
    • Holds OAuth tokens and a Yandex Cloud key
  • ycliCLI
    • Creates, edits and deletes issues, pages and forms in a live organization
    • Stores an OAuth token with write scopes
    • Can delete tasks, projects, boards and columns via tools flagged destructiveHint
    • The API key is permanent and grants company-wide access; the README explicitly warns about the risk if it leaks
  • yougile-mcpMCP server
    • Changes live company data in YouGile: tasks, users, roles
    • The API key grants company-wide access
    • Full CRUD over every company resource: projects, boards, tasks, users, departments and webhooks
    • The API key grants access to the whole company; the server does not scope permissions by role or project
    • Can delete company API keys and create or update users
    • Listens on all network interfaces (0.0.0.0) by default, not just localhost
    • Can obtain an API key from a YouGile account email and password
    • It downloads YouTube data in bulk in several parallel jobs, and on 403 errors the platform may restrict your access or IP
    • The --cookies-from-browser flag reads browser cookies that carry your session data
    • Transcript text goes to OpenAI or Gemini under your key for semantic search, chat and summaries
    • Write commands create, change, transition and delete issues, components and versions in the live Tracker
    • The CLI's policies do not protect against direct edits to the config file or raw service-account credentials in the caller's environment
    • Start with --read-only at login and allowed_queues set to the minimum needed, loosen only by deliberately recreating the profile
    • Write commands change issues, comments, attachments and Wiki pages in the live organization
    • There is a separate read and write allowlist, but it only guards calls through the CLI itself, not raw credentials in the environment
    • Start with a read-only allowlist (get, find, count) and keep write verbs in confirm mode
  • ytrCLI
    • Creates, edits and deletes issues, comments and worklogs in a live Tracker
    • Bulk operations affect many issues at once
    • Stores an OAuth token in a config file
  • yutuCLI
    • Publishes, changes and deletes videos, playlists, comments and subscriptions on the channel on behalf of the account owner
    • Keeps the channel's OAuth token in `youtube.token.json` and the app key in `client_secret.json`
    • The built-in agent sends the texts it reads to the model provider, and the HTTP mode opens port 8216
    • The agent runs shell commands and edits files, and without `--sandbox` it does so with your rights

How we set the access level

Methodology

The editors set the level and the reasons from the README, the way the tool runs and the permissions it needs. Reading project files the agent can already see does not raise the level. There is no automatic code audit yet.