Access levels of skills and MCP servers

Broad access

• BotHelp scenario MCP pluginPlugin
  • Edits a bot's live scenario graph: blocks, transitions, layout
  • Creates, repeats and cancels broadcasts to subscribers
  • Uses a captured operator session through a private, undocumented protocol
• Bright Data SkillsPlugin
  • The Bright Data key starts paid requests, and in the MCP URL it sits in plain text in the configuration
  • Scraping social networks and profiles collects personal data, and the skills bypass bot protection and CAPTCHAs
  • bdata login saves the key to disk and creates the cli_unlocker and cli_browser zones in your account
• Browser HarnessSkill
  • The agent drives your real Chrome and sees all open tabs, logins and cookies, not just one site
  • Through CDP it runs arbitrary code on pages, presses buttons, types text and uploads files under your account
  • With recordings on, screenshots and action traces with page content are saved to disk
• Browser Harness JSSkill
  • Attaches to your running Chrome and can read pages and cookies and type into sites where you are already logged in
  • The agent calls any protocol method, including file uploads and downloads, network interception and page printing
  • On first run installs the Bun runtime and keeps a background server on a local port until you stop it
• Browser MCPMCP server
  • Drives your browser with accounts already signed in
  • Can act on external sites on your behalf
• Browser MCPMCP server
  • Controls your browser with active logged-in sessions
  • Can act on external sites on your behalf
• Browser UseCLI
  • The agent acts in the browser on the user's behalf, including logged-in sessions
  • Runs arbitrary Python code with browser helpers
• BrowserOS neoSkill
  • The agent works in your accounts with saved logins, such as email, CRM and social media, and can send and change data
  • The `run` tool executes scripts in the browser, and the agent downloads and uploads files
  • Website pages remain untrusted data: they can contain instructions aimed against you
• BrowserTools MCPMCP server
  • Reads data from a logged-in browser session, including request bodies and storage
  • Requires an extension with DevTools access
• BrowserWingMCP server
  • Controls a real Chrome with your authenticated sessions and cookies
  • Can fill and submit forms and run arbitrary JavaScript on pages
  • The local HTTP service is unauthenticated by default
• Browserbase SkillsPlugin
  • Drives a browser and can act on sites on the user's behalf
  • Syncs cookies and works with authenticated sessions and an external cloud
• Bun-based Yandex Direct MCP with Unified Performance and Click.ru proxyMCP server
  • Creates and changes campaigns, including Unified Performance ones, with a real budget
  • Forecasts and changes bids, which directly affects spend
• Buoy MCPMCP server
  • Taps the screen, types and drives the simulator or phone, and with Pro changes storage values, Redux and Zustand state, the clock, location and app permissions
  • Reads network request bodies and storage values, which can hold tokens and personal data
  • The init command overwrites the buoy entry in MCP config files and copies a skill over matching files
• Burp Suite MCP ServerMCP server
  • Gives the agent access to intercepted traffic, including tokens and cookies
  • Can change Burp configuration when the checkbox is enabled
• BurpMCP-UltraMCP server
  • Sends real HTTP, WebSocket and fuzzing requests at targets inside the test scope, including the active scanner and Intruder
  • Injection, JWT and access-control checks are offensive techniques, legal only against systems you own or are authorized to test
• Butterbase SkillsPlugin
  • Through MCP the agent gets access to your Butterbase project: it applies schemas and access policies, deploys apps and moves them between regions
  • The skills work with Stripe Connect payments and service keys
  • Integrations send emails and SMS on behalf of your app
• CC SwitchCLI
  • Stores provider API keys and rewrites agent configs
  • The local proxy routes all agent traffic to the model through itself
• CCManagerCLI
  • Starts agent processes in a terminal and runs hook commands from settings on status changes and worktree creation
  • Creates, merges and deletes Git worktrees and copies ignored files such as `.env` and Claude Code session data
  • The auto approval mode answers the agent's permission prompts by itself
• CCPMSkill
  • Creates, comments on and closes GitHub issues via the gh CLI
  • Makes commits, sets up git worktrees and runs bash scripts
  • Launches parallel agents that modify the project's code
• CCSCLI
  • Stores OAuth tokens and API keys for several providers
  • Proxies all agent traffic through a local server
• CCXT MCP ServerMCP server
  • Can place orders and withdraw funds when enabled in config
  • Handles exchange API keys
• CDEK MCP and CLI: orders and trackingMCP server
  • It can create real delivery orders and waybills under live CDEK keys
  • The Client ID and Secret give full access to the business account
• CDEK MCP: calculator and HTTP serverMCP server
  • It can create real delivery orders and book couriers under live keys
  • The HTTP mode exposes the server over the network with no built-in description of external call authentication
• CLI Agent OrchestratorCLI
  • Launches coding agents that run commands and change project files
  • Starts a local server and coordinates several processes in tmux
• CLI Printing PressPlugin
  • The installer runs as `curl | bash`, performs `go install` and installs the skills globally into the agent
  • The skill records site traffic through a browser and calls service APIs under your authorization
  • The agent writes and runs generated Go code, and publishing a tool opens a PR to the shared library
• CLI-AnythingPlugin
  • Generates executable code and installs it to PATH
  • Installs third-party wrappers that run local programs
• CORALCLI
  • It launches several agents that edit code and commit in worktrees without your involvement at each step
  • It executes grader code and agent solutions on your machine; outside Docker the isolation of the agent from the grader is off
  • Each agent runs under your keys and sign-in, so a long run spends plan limits or API balance
• CORE: memory and tasks for Claude CodePlugin
  • Hooks send records of Claude Code sessions to app.getcore.me and use your account token
  • The platform reads email, calendar, trackers and chats, and the agents it starts run commands on the machine
• CTF SkillsSkill
  • The installer adds many system packages and tools
  • Skills run exploits, arbitrary shell commands and network attacks
• CTF Super HubSkill
  • Covers offensive security techniques: pwn, reverse, attacks on crypto
  • Can cause harm if used against third-party resources
• CTI ExpertSkill
  • A recon and OSINT tool: use only on your own assets or with explicit authorization, not against others' resources or people
  • It runs Python, installs tools and reaches many external services
• CTXMCP server
  • The agent reads, writes and searches files in every project from the configuration
  • Commands you describe run in the terminal with your user's rights: tests, migrations, deploys
• Caido SkillsSkill
  • The agent sends real requests to the target with the cookies and authorization headers taken from Caido history
  • Match and replace rules change the traffic that passes through your Caido
  • The PAT token gives access to your Caido instance, and its history can hold other peoples secrets
• CallMePlugin
  • Keeps the phone provider, OpenAI and ngrok keys in environment variables
  • Runs a local server, opens an ngrok tunnel to the internet and places paid calls
• CanopyMCP server
  • The agent's API key exposes the workspace's messages, files and tasks and lets it post as the agent
  • The app accepts inbound connections on ports 7770 and 7771 and links your computer to outside peers
• Canvas MCPMCP server
  • The token exposes grades, submissions and student correspondence, and educator tools change assignments, grades and pages
  • The agent can send messages and announcements in Canvas in your name
  • TypeScript execution runs code on your computer
• CashClawSkill
  • The skills invoice through Stripe, take payment and withdraw funds to the worker's wallet
  • Auto-accept takes jobs without human confirmation, and the skills send emails and messages to clients and work with WhatsApp
  • The data scraping skill crawls sites and compiles contact lists, and Stripe and HYRVE keys sit in the configuration
• CauraMCP server
  • Stores agent entries, flags personal data and can delete entries in bulk
  • Mandatory keystone rules override conflicting user instructions in a session
  • Shared memory carries one agent's entries to others, so a wrong entry spreads through the fleet
• CavemanSkill
  • The proxy sits in the traffic between the agent and the provider
  • The full installer wires Claude Code hooks and runs via curl | bash
  • The CLI sends anonymous telemetry by default
• CentralMind GatewayMCP server
  • Connects to a database with a login and password string and hands table data to the agent
  • In raw mode it runs the submitted SQL without a read-only check, and the database account sets the rights
  • While creating methods it sends the schema and sample rows to a third-party model provider
• CerseiCLI
  • The agent runs shell commands, reads and writes project files and goes to the web on your behalf
  • The --no-permissions flag removes confirmation prompts for dangerous actions
  • Anthropic or OpenAI keys pay for model calls, and project context goes to the chosen provider
• Charles MCP ServerMCP server
  • All responses come unmasked, so tokens, cookies and passwords from traffic reach the agent's context
  • The reverse_* tools replay requests with changed parameters and look for signature fields, which can affect live services
• ChatGPT CLICLI
  • The --agent mode runs shell commands and file operations at the model's discretion
  • The OpenAI or other provider key comes from environment variables, and query text with added context goes to the provider
  • Conversation history is stored on disk as plain text
• Chestny Znak MCP: marking codes and emissionMCP server
  • It can create code emission orders and submit application reports under the market participant's credentials
  • Errors in the unverified method map (verified: false) could lead to an incorrect call against a regulatory API
• Chrome MCP ServerMCP server
  • The agent acts in the browser under your accounts
  • Can read history, bookmarks and network traffic and inject scripts into pages
• CitadelPlugin
  • Runs step-by-step workflows and hooks that execute commands and modify files in the repository
  • Coordinates parallel agents working on the codebase
• Claude AdsPlugin
  • Can work with live ad accounts and budgets
  • Stores and uses ad platform API credentials
  • Runs Python scripts and a browser for page capture
• Claude CodeCLI
  • The agent runs shell commands and modifies files
  • The installer downloads and runs a script
  • Code is sent to Anthropic models
• Claude Code ActionPlugin
  • Writes to the repository: creates branches, commits and comments
  • Handles API secrets in GitHub Actions
• Claude Code Apple SkillsPlugin
  • Generators write code into your project: networking, purchases, authentication and persistence layers
  • The `asc.py` helper with your App Store Connect key can change in-app purchase prices and URLs, though by default it only prints the request and writes after `--apply`
  • The script needs a private key with the App Manager role, stored on your disk