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

    • The token is derived from the Yandex account Session_id cookie
    • An OAuth token with broad access is stored in .env for about a year
    • The script connects to devices on the local network
    • Creates and edits issues, statuses and worklogs in a live Tracker
    • Stores the OAuth token in config files
    • Creates and modifies issues, comments and statuses in the live Tracker
    • Handles organization OAuth and IAM tokens
    • Write actions create, update and transition issues in the live Tracker
    • It supports bulk operations and automation mutations (triggers, autoactions, macros)
    • Start with TRACKER_READ_ONLY=true until you trust the server's behavior
    • Creates, edits and deletes issues, boards, columns and sprints
    • Bulk operations change status or move many issues at once in a single call
    • Creates issues, comments and goals and bulk-changes Tracker statuses, projects and queues with a user's token
    • The Tracker token is stored in the server settings in Yandex Cloud and grants access within its owner's rights
    • Can create, change and delete issues, entities, comments and attachments in the real Tracker
    • 146 tools grant broad write access, including bulk operations
    • Creates, updates and closes issues in the real Tracker when confirmed=true
    • Scheduled reports can be sent to an external Telegram bot with real issue data
    • Stores the OAuth token and API key in .env
    • Without TRACKER_READ_ONLY it changes real Tracker issues on behalf of the authorized user
    • Its own OAuth provider stores and encrypts user tokens on the server
    • No README, which makes config review harder before production use
    • Typed and raw tools create and change issues, queues and other Tracker entities
    • tracker_raw_request opens access to any v3 API method within the v3 domain
    • Can remove a site, sitemaps and feeds from Webmaster
    • Recrawl submissions consume a limited quota
    • Runs a server via npx with OAuth sign-in to your Yandex Webmaster account
    • The recrawl_submit tool sends pages for recrawl and spends the site quota
    • The other six tools only read statistics
    • Write tools change campaigns, bids, ads and goals in Direct and Metrika
    • It can upload offline conversions and CRM orders that affect the client's statistics
    • Start with a read-only token or the Direct Sandbox until you have verified the server's behavior
  • Yandex Wiki MCPMCP server
    • In full mode it creates and overwrites corporate Wiki pages
    • Uses an organization OAuth token
    • A single token opens mail, disk, calendar, ads and the task tracker at once
    • The Direct plugin, once activated, can change ad campaigns and spend budget
    • The token is stored on disk in a file rather than a system secrets store
    • Can move, overwrite and permanently delete files on Disk
    • Stores a write-scoped OAuth token in the client config
    • The sampling skills drive your signed-in browser for ChatGPT, DeepSeek or Doubao and send dozens of questions in a row from your account
    • Python and Node.js scripts write reports, tables and screenshots to folders on disk
    • `yao-geoflow-cli` operates a GEOFlow system through CLI, API and admin: it can change materials, tasks, users and tokens, and destructive actions require confirmation
  • YooKassa MCPMCP server
    • Creates payments, refunds and payouts with real money
    • Requires the shop secret key
    • Recurring charges run without the buyer's involvement
    • The key gives the agent access to the shop's payment data, including amounts, statuses and refunds
    • Start with a test-shop key and switch to the real one only after checking how the agent behaves
    • The project is in the payments domain: even without write tools, the agent works with credentials and money-acceptance logic
    • Diagnostics reads test-shop payment, refund and receipt data using a secret key
  • YouGile MCPMCP server
    • Creates and changes tasks, boards and projects in the workspace
    • Can invite and remove company users
    • Requires a company API key
    • All traffic between employee agents and the company's YouGile passes through Indalo's third-party hosting
    • A new employee gets the company's default permissions, which may be unrestricted until an admin configures them explicitly
    • Full CRUD including deleting projects, boards and columns via archiving
    • Uploaded files become publicly reachable by URL with no authentication
    • Can create, update and delete tasks, projects and users in YouGile
    • Stores the account email and password in config instead of a scoped API key
    • The default admin role grants full access to the entire API, including employees, projects and webhooks
    • Permissions only narrow what the key already allows in YouGile, they don't replace the service's own protections
    • yougile_request gives access to nearly any YouGile API endpoint with no read/write distinction
    • The yougile_create_api_key tool accepts a YouGile account password through MCP
  • YzelMCP server
    • Creates and updates 1C documents and CRM and marketplace data
    • Stores credentials for several production systems on one machine
    • The project is alpha and mostly tested against mocks
    • The full Zabbix API: editing and deleting hosts, templates and triggers
    • The Zabbix token opens monitoring of the whole infrastructure
    • The read-only mode has to be turned on by you
    • Performs actions in external apps on the user's behalf
    • Data passes through a third-party platform
  • ZeroCLI
    • The agent writes files and runs shell commands, and autonomous modes without prompts are a separate choice
    • Keeps provider keys and subscription logins in user settings and sends project content to the chosen model
    • Plugins and hooks run external commands on beforeTool, afterTool and session events
    • If the compiler is missing, the skill downloads an install script from the network and runs it in the shell
    • The agent runs built programs with `zero run`, with access to your machine's input and output
    • The author warns about experimental status and possible security holes
    • Several agents edit code and run commands
    • Spends more tokens than one agent
    • Downloads Chrome and runs a browser that executes page JavaScript
    • Can use saved sessions, cookies and credentials
    • Page content may contain prompt injection
    • Sees and presses any element in any app, can force-quit them, type text and send messages in your name
    • Reads and writes the clipboard and reads notifications, which can hold codes and private messages
    • Asks macOS for Accessibility and Screen Recording, and the permission goes to the whole terminal
    • The manager launches agents and shells on your machine, including in separate git worktrees
    • Through MCP, agents gain the ability to launch other agents and message them
    • The install script downloads and installs a binary from a release
    • By default the skill edits code, commits, pushes to your branch and writes replies under your GitHub account
    • The tool reads a token from `GITHUB_TOKEN`, `GH_TOKEN`, `.env.local` or `gh auth token`
    • Takes a session token from Slack Desktop or the browser and acts in Slack as you
    • Can send, edit and delete messages, invite people, create channels and run Slack workflows
    • Reads direct messages and channels, and the content lands in the model's context; safe mode `AGENT_SLACK_SAFE_MODE=1` limits the damage
    • With your consent onboard writes hooks and MCP entries into client settings, and an environment variable for Claude Code
    • Hooks fire on every tool call of the agent, while sync and the API stay running in the background on 127.0.0.1
    • The tool reads agent session files, which can contain paths and commands from your projects
    • The project's source code goes to the chosen model provider unless you use local Ollama
    • agentgg init stores the provider key on disk
    • Agents spend tokens on every scan and write reports to the directory given in -o
    • Loop mode runs agents that change code in worktrees
    • Installs hooks and scripts into the project
    • Deploys agents to Google Cloud, provisions infrastructure and handles secrets
    • Runs build, test and deploy commands in the project
  • agmsgSkill
    • Writes hooks into the project's `.claude/settings.local.json` and the Codex settings, and in `monitor` mode puts a shim in place of launching `codex`
    • `spawn` opens new agent processes in a terminal, and the spawn options file lets you pass them flags such as `--dangerously-skip-permissions`
    • The whole conversation is stored as plain text in SQLite under `~/.agents/skills/agmsg/` and, with remote sync on, goes to your own server
  • agtxCLI
    • Runs agents that write and delete files and execute commands in working copies, and the oneshot skill merges the result into the local base branch by itself
    • The mobile board opens reading of tasks and agent screens and typing into an agent session, which is arbitrary command execution on your computer
    • Contacts GitHub once a day for a new release, and the install script and the update download a binary
    • The plugin ships a .mcp.json with remote Slack, HubSpot, Klaviyo, Notion, Canva, Figma and other servers
    • Once you sign in to those services the agent can post to Slack and change CRM data and email campaigns
    • The install.sh script copies the skills into the global agent folder
    • Through a browser tool the skill can press buttons on sites where you are already signed in
    • Generation on paid platforms spends credits, although the skill stops before payment and publishing
    • Downloaded results and images you upload pass through third-party generation services
    • Creates, modifies and deletes records and tables in an external service
    • Uses an Airtable access token
    • Creates and updates deals, contacts and tasks in a live amoCRM account, including in bulk
    • Holds a long-lived token with access to customer data
    • Can route account events to an external URL via webhooks
    • Creates and changes leads, contacts and companies in a live amoCRM account
    • The integration token grants full access, not read-only
    • Creates and bulk-updates leads, contacts and companies in the live CRM
    • Accepts and rejects incoming requests
    • Stores OAuth tokens with access to the customer base
    • Creates and updates leads, contacts and companies
    • Stores the OAuth client_secret and refresh_token for auto-refresh

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.