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

  • Meta Ads MCPMCP server
    • Changes a real Meta ad account and spends real budget
    • Requires account access and ad platform tokens
    • Generates executable code: a CLI, an MCP server and config files for Claude Code, Codex, Copilot and other hosts
    • The signing and publishing subcommands deal with a signing key, IPFS and your npm account, and the plugin states it validates secrets through GCP Secret Manager
    • Network, shell and file-write grants in the generated MCP can be turned on in the policy, after which the agent has them in full
  • MetaMCPMCP server
    • Stores secrets and launches the connected MCP servers
    • Exposes tool access over the network
    • The agent opens, changes and closes trades on a live account, so it handles the user's money
    • The account password sits in a configuration or an .env file, and the network mode checks no access and listens on all interfaces
    • The agent reads and writes files and runs shell commands in the current folder
    • Subscription sign-in and provider keys are saved in the ~/.metis/agent folder on your computer
    • The skill embeds into Codex, OpenCode or DeepSeek Harness and changes how those agents work
  • Metricool MCPMCP server
    • Publishes posts and replies on social networks on behalf of your brand
    • Runs ad campaigns and can spend the ad budget
    • Requests to social networks and analytics are processed by the external Metricool service
    • Reads mail, files and the calendar and can change them
    • Rights depend on the Azure app permissions
    • The server is unofficial
    • The skills describe DDL, DML and COPY INTO, so the agent can change warehouse schema and data
    • Work runs under your Azure sign-in with a token for api.fabric.microsoft.com
    • The plugin attaches remote Fabric MCP servers, including one that executes SQL queries
    • The server runs INSERT, UPDATE and DELETE as the database user, and there is no read-only mode
    • The database password sits in the client configuration, and query results, including personal data, enter the agent context
    • Reads Microsoft 365 work data: mail, documents, meetings
    • Access to tenant data requires administrator consent
    • The set is in Public Preview
  • MikroTik MCPMCP server
    • The agent changes router configuration: firewall and NAT rules, users, VLANs, routes
    • The SSH login and password live in environment variables or the client config
    • In HTTP mode the server listens on all interfaces by default, so restrict access with the allowed-ips option
    • The plugin stores the conversation at the end of every turn, and messages can contain personal data, keys and work secrets
    • In cloud mode sends messages to an external server and keeps the API key in `~/.mindmemos/settings.json`
    • The `memory delete`, `memory update` and `skill push` commands change and delete data on the server
    • The agent writes files and runs commands in your project, although edits can be approved before writing
    • The API token is stored in ~/.mini-code/settings.json
    • Connected MCP servers and web requests widen the agent's access
  • Mission ControlMCP server
    • Stores API keys and credentials for controlling several agents
    • Lets callers register agents and dispatch tasks to them programmatically
    • The README explicitly warns to keep the instance on a trusted network without a TLS reverse proxy
  • MoLingMCP server
    • The terminal tool runs system commands as your user
    • File tools write and merge files, so an agent mistake can overwrite something you need
    • The browser tool controls your Chrome and sees open pages
  • Mobile MCPMCP server
    • Controls the device: installs and removes apps, types text
    • On a real device the agent can reach personal data and accounts
    • Highly sensitive financial data
    • The session key equals the password
    • An unofficial library, breakages possible
    • Without --readOnly it can delete documents, collections and databases
    • Can create and modify Atlas clusters, affecting costs
    • Handles connection strings and secrets
    • The agent gets write access to project assets: Blueprints, materials, configuration, builds and play in the editor
    • Port 9316 is open on all network interfaces, and other devices on the same network can reach the server directly
    • Voice and music through third-party providers
    • The models are large and rendering is heavy
  • MoySklad MCPMCP server
    • Creates payments, cash orders and invoices in a live account
    • Creates warehouse documents and changes prices
    • The token grants access to all of the MoySklad user's data
    • The app can create and edit documents, products and orders in the company's accounting system
    • Data passes through a third-party partner's infrastructure rather than MoySklad's own, which is worth weighing when choosing access rights
    • Creates real orders in MoySklad for whatever amount the agent specifies
    • Reads counterparties with INNs, phone numbers and addresses
    • A single server holds tokens for several client Kommo accounts at once
    • The Antigravity connection method runs JS code fetched live from GitHub, with no version check
    • Writes and changes tasks and data in the live Pyrus account across several tenants
    • Stores credentials and a call audit trail in its own database accessible to the server
    • Runs arbitrary SQL, including INSERT, UPDATE and DELETE in the database
    • Handles database credentials and can change data when granted write access
    • A network installer sets up system software
    • Manages autonomous agents and inference credentials
    • Every generation goes to the Gemini API under your key and spends your money
    • Install via `/init` clones the repository, installs dependencies and writes the key to `~/.nano-banana/.env`
    • Agents run commands and access mounted folders
    • Connects to messengers and can send messages
    • Generates and runs experiment code
    • Calls external models and catalogs with a key
    • Plugin installation registers hooks on every prompt, tool call and session end in `~/.claude/settings.json` and injects results into the system prompt
    • Stores and indexes past session text that may contain secrets and code; the hosted gateway accepts such data over HTTPS with a token
    • Runs a persistent local daemon and writes files under `~/.claude`; tokens with read access across all projects need a separate flag
  • NelmCLI
    • Applies and deletes Kubernetes resources
    • Handles secret encryption keys
    • Runs write Cypher queries against the database
    • Can create and delete Aura cloud instances
  • Neon MCP ServerMCP server
    • Can create and delete projects and branches
    • Runs SQL against a cloud database
    • Deploys sites and changes project settings in an external service
    • Manages environment variables and secrets
    • Hooks into Claude Code, Cursor and Windsurf settings and captures every prompt before sending
    • Stores prompts in ~/.nexpath and sends parts of them to OpenAI or the Nexpath service for classification
    • Keeps the OpenAI key or Nexpath token in the system keychain or in config.json
    • The gateway holds LLM provider keys and MCP server tokens and passes every agent tool call through itself
    • STDIO servers from the configuration start as child processes with an arbitrary `cmd` command, environment variables and a working directory
    • Without OAuth2 the gateway accepts any string in place of a key, so the port has to be kept closed: by default it listens on 127.0.0.1:8000
    • The server launches a Node.js script on your machine with debugger access to the process
    • The evaluate_expression tool runs arbitrary JavaScript in the paused process
    • Variables and console output may contain secrets, which then reach the agent and the model
  • NorthCinderMCP server
    • It touches purchase checkout, though each one needs a separate approval
    • It goes online for store and product data
    • Executes changes in connected ad accounts, which directly affects budgets
    • Modifies content and settings in WordPress and data in the GoHighLevel CRM
    • Connects over OAuth to third-party platforms and operates on production accounts
    • Automates a browser logged into your Google account and stores the session
    • Installs Chrome and Python dependencies
    • The project is archived and receives no fixes
    • The agent creates, edits and archives pages, databases and comments on your behalf
    • A personal token opens every page you can see, not a few chosen ones
    • Writes and updates real Notion pages and databases
    • Handles a workspace access token
    • Modifies workspace pages and databases
    • Exposes connected page content to the model
    • Creates and edits Notion pages and database rows with the rights of your account
    • Reads any workspace content that your sign-in can reach
    • Connects a remote MCP server that receives requests and page content
    • The agent runs arbitrary PHP and WP-CLI, writes and deletes files and works with the site database with administrator rights
    • Read mode limits only what Novamira offers, while access through administrator credentials stays full
    • Sends user documents to an external cloud API
    • Operations spend paid credits
    • A key with Trade permission lets an agent place, amend and cancel orders and launch trading bots with real money
    • The tools expose your account's balance, positions and operation history
    • API keys sit in the ~/.okx/config.toml file on your computer, and access to it equals access to the account
    • Generates and fires attacks at a target, use only on your own systems and with permission
    • Needs a model provider key, keep it out of chats and repositories
    • Consumes provider tokens and money on runs
    • Grants access to write operations via the Seller and Performance API, including prices and orders
    • Layered protection reduces but does not eliminate the risk of a mistaken call

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.