# Plugins for an AI agent Telegram bot: installation and checking > Where a plugin is installed, how it differs from a skill and MCP, how to reload the agent, and how to check commands from Telegram. Source: https://ailoot.ai/blog/en/telegram-ai-agent-plugins/ Canonical: https://ailoot.ai/blog/en/telegram-ai-agent-plugins/ Language: en Publisher: Ailoot (https://ailoot.ai/) Translation: guide-151 Revision: sha256-01735e5da25f6aa5ee8e8acdce82d48bc2e0bd22435e330d6f2224cb913bf47c My agent runs on a server, and I give it commands from Telegram: I write in the chat, it reads files, runs tasks, and sends me the result. At some point, you hit a ceiling: the basic bot can do very little, and you want it to review code, commit to GitHub, or compile a scheduled summary. The answer is plugins. People search for “how to install a plugin in Telegram” precisely for this. They are not talking about a sticker store. They want to know how to give a bot new abilities. Below, step by step: what a plugin is, where it is actually installed, how to install one in five minutes, and how not to bring malicious code into your agent. This description is based on the official documentation. [Practical information about working with Ailoot is in the documentation](https://ailoot.ai/docs/). ## What is an AI agent plugin, in simple terms?
Key takeaway A plugin is a ready-made set of abilities that you add to an agent with one command. It contains skills, separate helpers, automated actions, and connections to external services. Install a plugin, and the bot gets new commands it did not have out of the box. Remove the plugin, and they disappear. It is like installing an app on a phone, except the app is installed on your AI agent.
Let’s break it down. An AI agent out of the box can do the basics: read files, write code, and answer questions. A plugin is a folder of additional abilities that the agent picks up and starts using. The official Claude Code documentation describes a plugin exactly this way: it extends the agent with skills, separate helper agents, automated actions, and connections to external services. Here’s a simple example. Install a plugin for working with GitHub, and the agent gets commands to create a commit, open a pull request, and review the change history. Install a review plugin, and the agent checks your code for errors and security holes using a predefined checklist. You did not configure any of this by hand: the plugin author put it all together, and you just connected it. One more important thing: a plugin does not teach the agent anything “forever.” It is a removable module. Connect it today, disconnect it tomorrow, and the bot is back to the set of abilities it had before the plugin. That makes plugins convenient to try: pick one up, run it on a real task, and if it does not work for you, remove it without a trace. ## How is a plugin different from a skill, MCP, and a subagent?
Key takeaway A skill is one ability, a separate helper agent takes on a larger subtask, and a connection to an external service gives the bot access to another program. A plugin is a box containing all of these, along with a description file. Install one box and get a whole set of abilities at once, instead of assembling them one by one.
The terms around agents can be confusing, so I’ve summed up the differences in a table. It also answers half the questions newcomers have. | What it is | What it does | How it is installed | |----|----|----| | Skill | One specific ability: “create a commit in our style” | Put an instruction file in the agent’s folder | | Helper agent | Takes on a larger subtask in a separate conversation without using up your context | Describe it in a separate file | | Automated action | Triggers automatically on an event: before saving, after a response | Add a rule in the settings | | Service connection (MCP) | Gives the bot access to an external program: GitHub, a database, a browser | Configure the address and access key | | Plugin | A box containing all of these, assembled and packaged | One installation command | In the Claude Code world, a connection to an external service is called MCP. It is a standard that lets the agent reach GitHub, a database, or a browser. The first four rows of the table are separate building blocks. Look at the bottom row. A plugin sits one level above the other four. It brings together a skill, a helper, an automated action, and MCP in one box. The author put everything together, added a description file, and published it. You get a ready-made whole with one command instead of assembling four pieces separately. That gives you a simple rule for choosing. If you need one small ability for your task, write the skill yourself; it takes fifteen minutes. If you need a ready-made set from people who have already solved your problem, get a plugin. I have a separate guide on [how to choose skills for Claude Code](https://ailoot.ai/blog/en/choose-claude-code-skills/), including how not to overload your agent with extra skills. ## What does “install a plugin in Telegram” mean, and what does Telegram have to do with it?
Key takeaway The plugin is installed in the agent that responds to you through Telegram, not in Telegram. The messenger itself is a remote control: you use it to give the bot commands, while the agent runs on a computer or server. So “install a plugin in Telegram” really means “install a plugin in the agent and keep controlling it from the chat.”
This is where a common confusion comes in. Telegram itself has no plugins: the messenger does not store the bot’s abilities. It passes your messages to the agent and displays its replies. The brain is somewhere else—on your laptop or on a server running an agent such as [Hermes Agent](https://ailoot.ai/blog/en/hermes-agent-install-setup/), [OpenClaw](https://ailoot.ai/blog/en/openclaw-install-telegram/), or Claude Code connected to the chat. This leaves two ways to install a plugin, and both work: 1. **Right from the chat.** Write a plain-language request to the agent in Telegram: “install a GitHub plugin for yourself.” The agent installs it on its own machine and replies that it is ready. This is convenient when the server is far away and you are on your phone. 2. **In the terminal.** Log in to the machine where the agent runs and install the plugin there with a command. This is more reliable the first time: you can see all the output and won’t miss anything. Which method to choose depends on whether you trust the bot to install extensions on its own without supervision. I installed my first plugins by hand in the terminal so I could see exactly what was being connected. Once I got the hang of it, I started trusting the agent to install them itself from the chat. I’ll show you both ways below. ## How do you install a plugin in an agent in five minutes?
Key takeaway First, connect a plugin marketplace, then install the one you want by name, and then reload the agent so it picks up the new plugin. In Claude Code, that takes three commands: add a marketplace, install a plugin, and reload the plugins. Hermes Agent and OpenClaw follow the same logic, with different command names.
I’ll show you Claude Code because the process is described word for word in its documentation and is easy to repeat. The commands are the same in the terminal and when you send them to the agent in Telegram. 1. **Open the plugin list.** The `/plugin` command opens a panel directly in the terminal, with a tab showing all available extensions and a directory on claude.com/plugins. This is a storefront where you can browse before installing. 2. **Connect the official marketplace.** Enter `/plugin marketplace add anthropics/claude-plugins-official`. A marketplace is a directory from which the agent downloads plugins. The official one is run by Anthropic. 3. **Install the plugin you need.** Use the plugin name followed by the marketplace name, separated by an at sign: ```bash /plugin install github@claude-plugins-official ``` 4. **Reload the plugins.** If after installation the bot says “Run /reload-plugins to activate.”, run `/reload-plugins`. Until you reload them, the new commands will not appear in the current conversation. If you get a warning that reloading will reread the conversation, repeat the command as `/reload-plugins --force`. That’s it. The agent now has the plugin’s commands. In Claude Code, they use the plugin name as a prefix, for example `/commit-commands:commit`, so you can immediately see that the command came from a plugin. When you send a command to the bot in Telegram instead of using the terminal, Claude Code also has a standalone installation command without the interactive panel: `claude plugin install`. The documentation notes one detail: this command does not load the plugin into an already open conversation. The agent will pick it up at the next launch or after `/reload-plugins`. In other words, after installing, you need to restart the bot or you won’t get the new commands. In Hermes Agent, extensions are enabled and disabled with the `hermes tools` command, while skills live in their own folder. OpenClaw works similarly: connect a set and restart the bot. The names differ, but the order is the same: connect a source, install by name, and restart the agent. Plugins are one building block of context engineering: you decide which abilities the agent keeps at hand before the first request. ## Where should you get plugins, and which ones can you trust?
Key takeaway There are three sources with different levels of reliability: Anthropic’s official marketplace, a vetted community marketplace, and any third-party directories on GitHub. The vendor itself runs the first; plugins in the second have passed an automated check and are pinned to a specific version; the third is at your own risk. The further you get from the official source, the more carefully you should check what is inside.
A plugin is someone else’s code, and it gets the same permissions as your agent: files, terminal, and access to services. That is why the source matters more than a nice description. Here are the sources, from most reliable to most questionable. | Source | Who is responsible | How to connect | |----|----|----| | Official marketplace | Anthropic maintains the directory itself | `/plugin marketplace add anthropics/claude-plugins-official` | | Community marketplace | Third-party plugins have passed an automated check, and each is pinned to a version | `/plugin marketplace add anthropics/claude-plugins-community` | | Vendor examples | Anthropic, sample plugins | `/plugin marketplace add anthropics/claude-code` | | Third-party directory | The author on GitHub; nobody has checked it | Use the repository address, at your own risk | To install from the community marketplace, use a similar command, but with a different marketplace: `/plugin install plugin-name@claude-community`. The documentation specifically notes that these plugins have passed an automated security check and are each pinned to a specific version of the code. That means you’ll get exactly what was checked, without any later changes slipped in by the author. Before installation, Claude Code shows a “Will install” section: a list of the commands, helpers, skills, automated actions, and connections the plugin will add. Read it. This is your chance to spot a notes extension that is asking for terminal access for some reason. The official documentation puts the rule plainly:

Make sure you trust a plugin before installing it. Anthropic does not control which connections, files, or other code are included in a plugin, and cannot guarantee that it will work as intended. Review each plugin’s page.

— Claude Code documentation, https://docs.claude.com/en/docs/claude-code/discover-plugins
In plain terms: the official and vetted marketplaces are for peace of mind; use a random GitHub repository only after reading what is inside and understanding why it needs the permissions it requests. Trust in third-party code goes beyond plugin installation, and I cover it in my article about [vibe-coding security](https://ailoot.ai/blog/en/vibe-coding-security/). ## How do you check that the plugin is installed and working in Telegram?
Key takeaway Ask the agent for a list of its commands or open the plugin panel. If the plugin’s commands are in the list, it is connected. Next, give it a small test task right in the chat and see whether the bot takes it on using its new ability. If nothing appears, you may have forgotten to reload the plugins or connected the wrong marketplace.
Spend a minute checking: seeing “installed” does not yet mean that the bot has actually picked up the plugin. Follow these steps: 1. **Check the list.** Open the `/plugin` panel again or ask the agent to show its available commands. The new plugin’s commands should appear in the list, prefixed with the plugin name in Claude Code. 2. **Give it a test task.** Choose something safe, not a live task. If you installed a GitHub plugin, ask it to show the latest commits. If you installed a review plugin, give it a small file. Look for one thing: does the bot take on the task using that new ability? 3. **Check from the same chat you use for work.** If you control the agent from Telegram, send the test command in Telegram. Sometimes the plugin is visible in the terminal, but the bot in the chat has not reloaded it yet. The most common reason a plugin is installed but its commands do not work is that you skipped the reload. The installation command succeeded, but the agent in the open conversation is still using its old set of abilities. `/reload-plugins` helps; if that does not work, fully restart the bot. The second reason is installing from a different marketplace: the plugin name matches, but the directory is wrong, and the agent correctly says that the plugin is not in the connected marketplaces. If the bot replies in the chat but does not understand the new command, it does not mean the agent is broken or that you did anything wrong. Almost always, it is one of two things: it has not reloaded the plugins, or you connected the wrong marketplace. Both take a minute to fix. ## How do you update and remove plugins without cluttering the agent?
Key takeaway Update the marketplace with one command; only then will fresh plugin versions become available. Removing a plugin is just as easy as installing one: use the same panel. Keep only the plugins you actually use connected: every extra plugin adds commands to the bot and fills its context.
There is a non-obvious detail about updates. When the author releases a new version of a plugin, it does not arrive on its own. First, update the marketplace directory with `/plugin marketplace update`, and only then install the fresh version. Until you update the marketplace, the agent considers the version it downloaded last time to be current. This is intentional: other people’s changes will not be slipped in after the fact without your knowledge. Remove a plugin through the same `/plugin` panel you used to install it: find the plugin in the installed list and remove it. It leaves no trace in the agent, and the commands disappear as soon as you reload. That is exactly why plugins are not scary to experiment with: install a bunch, keep the two that work, and remove the rest. And keep things tidy. Every connected plugin adds commands for the agent to keep in mind and uses part of its context window for their descriptions. Five plugins you do not use slow the bot down and confuse it for no reason. The rule is simple: if a plugin has not been useful in two weeks, remove it. You can restore it with one command. ## Which plugins would I install first?
Key takeaway Start with three areas: working with GitHub, code review, and commits in a consistent style. Almost everyone building a project and managing an agent from chat needs these. Add the rest for specific tasks. There is no need to install plugins in advance: unused plugins only slow the bot down.
I won’t give you a top 100 list—this deserves a separate guide, and I already have a top 10 Claude Code plugins article. Here are three areas I recommend for beginners who manage an agent from Telegram. 1. **Working with GitHub.** A plugin that gives the agent commands for commits, branches, and pull requests. Write to the bot in chat, “save the work and push it to the repository,” and it does it. This removes the main worry for people vibe-coding without experience: fear of losing code. 2. **Code review.** A plugin with a checklist for checking errors, security holes, and weak spots. It is especially useful with Telegram: send a piece of code in the chat and get a review without opening an editor. 3. **Commits in a consistent style.** A small plugin that makes the agent format commits according to one template. It may seem minor, but after a month the project history looks tidy, like a team’s, instead of turning into a mess. From there, choose plugins for specific tasks. Working with a database? Get a database plugin. Compiling summaries? Get a scheduling plugin. The general rule is the same: start with a task, then find a plugin for it. Installing plugins in advance, “just in case,” is a sure way to drown the bot in commands you never use. ## What else do people ask about plugins in a Telegram agent?
Key takeaway Quick answers to common questions: plugins are mostly free, work with both cloud and local bots, and a “Telegram plugin” from a sticker store is a different thing from an AI agent extension. You can see dangerous permissions before installation, and undo an installation in a minute.
### Frequently asked questions Are agent plugins paid? The plugins in the official and community marketplaces are free: they are open source, and there is no charge to connect them. The model connected to the agent may be paid, as may an external service if the plugin uses one. But you do not pay just to install a plugin. Can I install a plugin if the bot runs on a server? Yes, and that is the most common case. Install the plugin with the same command, either by logging in to the server’s terminal or by giving the command to the agent right in Telegram if you trust it to install without supervision. After installation, you need to restart the bot or tell it to reload the plugins, or it will not pick up the new abilities. How is an AI agent plugin different from a Telegram plugin? They are different things that people confuse because they share the word “plugin.” Telegram itself has no bot extensions: the messenger only passes messages along. The plugin is installed in the AI agent that responds to you through Telegram, and the agent runs on a computer or server. It is not inside the messenger. How can I tell whether a plugin is trustworthy? Check the source and the list of permissions. The vendor runs the official marketplace, while community plugins have passed an automated check and are pinned to a version. Before installation, the agent shows you exactly what the plugin will add: if a notes extension asks for terminal access, that is a reason to be cautious and not install it. What should I do if the plugin does not work after installation? Almost always, the reason is one of two things: you did not reload the plugins after installation, or you connected the wrong marketplace. Tell the agent to reload the plugins, and if that does not help, restart the bot completely. Check from the same chat where you work: the plugin may be visible in the terminal, while the bot in Telegram has not picked it up yet.
### Sources - [Claude Code, how plugins work](https://docs.claude.com/en/docs/claude-code/plugins) - [Claude Code, finding and installing plugins](https://docs.claude.com/en/docs/claude-code/discover-plugins) - [Claude Code, plugin marketplaces](https://docs.claude.com/en/docs/claude-code/plugin-marketplaces) - [Official Claude Code plugin directory](https://claude.com/plugins) - [Community plugin marketplace on GitHub](https://github.com/anthropics/claude-plugins-community) - [Hermes Agent, documentation](https://hermes-agent.nousresearch.com/docs/) - [Hermes Agent, connecting Telegram](https://hermes-agent.nousresearch.com/docs/user-guide/messaging/telegram) - [OpenClaw, official website](https://openclaw.ai/) - Guide: top 10 Claude Code plugins - [Guide: how to choose skills for Claude Code](https://ailoot.ai/blog/en/choose-claude-code-skills/) - [Guide: vibe-coding security](https://ailoot.ai/blog/en/vibe-coding-security/)
[Open Ailoot](https://ailoot.ai/). ## Questions about Plugins for an AI agent Telegram bot: installation and checking ### Why does an installed plugin still not work through Telegram? The running agent may not have loaded it yet. Reload plugins or restart the bot, then check the commands and run a harmless test from the same Telegram chat; installation output alone does not prove the active session picked it up. ### Where should the plugin be installed if the bot runs on a server? Install it in the agent on the machine that runs the bot. Telegram passes messages and displays replies; it does not store the agent’s extensions. Verify the source and what the plugin will add before installing it there. ## Ailoot · The company and its project portfolio Ailoot combines established operating experience, a large client base, stable service and ongoing KPI control. The company and its project portfolio: millions of customers; monthly sales in the billions. Ailoot brings that operating experience to API access for AI features in customer products, agents, applications and automations.