The extensions marketplace

The extensions marketplace is the catalog of everything you can add to the app. It does not ship inside the program: it is published from the web, so new extensions show up without you updating anything. You open the gallery and they are already there.

The five kinds of extension

MCP servers

They give the assistant new tools: reading your email, querying a database, driving a browser. They are the plug between the assistant and an outside service.

Skills

They teach the model to do one specific thing, and do it well: how to write your reports, how to review a code change, how to prepare an invoice. They do not add tools, they add craft.

Plugins

Packages that bundle skills and MCP servers into a single install, so you don’t have to add them piece by piece.

Agents

Agents already configured and ready to work. You install one and it already knows its role, its instructions, its tools and which model to think with.

Teams

Several coordinated agents, with a coordinator that hands out the work among them.

Packaged agents are the gem

Almost nobody offers agents already assembled. The usual thing is to be handed a tool and left to decide the role, write the instructions, pick permissions and guess the right model. Not here: you install an agent and it already knows who it is and what it does.

A team goes one step further. A social media team, for instance, comes with a campaign coordinator and one agent per network: one for Instagram, one for X, one for LinkedIn. The coordinator hands out the work and each agent writes in the voice of its network.

How to install an extension

1. Open the extensions gallery from the app’s side panel.

2. Pick the tab for the kind you are after: MCP servers, skills, plugins, agents or teams.

3. Search by name or by what you want to get done. You can also paste the identifier given by the public extensions page: the gallery recognizes it and takes you straight to the listing.

4. Open the listing and read it: what it does, who publishes it, which credentials it asks for and which permissions it declares.

5. Press install. The extension is available right away; MCP servers and skills load on their own.

The extensions gallery: one tab per kind, with the search box on top.

The extensions gallery: one tab per kind, with the search box on top.

What you see before installing (and why it matters)

Every extension’s listing shows, before you touch anything:

• Official — published by the Context Code team.
• Community — published by someone else. It may be excellent, but it isn’t ours.
• Verified — reviewed by hand. The badge is granted by the Context Code team, never by the author.
• Credentials it asks for: the names of the keys or accounts you’ll have to connect for it to work.
• Permissions it declares: what it is being authorized to do.
• Checksum: the package’s fingerprint, so you can check that what gets installed is exactly what was published.

Caution

A Community extension deserves a look before you install it. Read who publishes it, which permissions it asks for and why it says it needs them. If something doesn’t add up — it asks for terminal access for a task that clearly doesn’t need it, say — don’t install it.

Sensitive permissions

If an extension declares computer-use, terminal, remoto (remote) or instalacion-persistente (persistent install), the app warns you before installing and puts the list in plain sight. These are permissions that hand over control of your machine, and the decision is yours, with the facts in front of you.

Installing an agent does not run code

Worth being clear about: an agent brings instructions and configuration, not a program that starts on its own. Installing it launches nothing on your machine.

What an agent can do is ask for tools and permissions — an MCP server, terminal access, a credential — and all of that is visible on the listing before you install.

What a packaged agent brings

• Who it is: its role and its instructions, written by whoever published it.
• What it thinks with: a recommended model plus alternatives, among them always a local one with Context Forge so you can use it without spending.
• What it needs: tools, permissions, skills and MCP servers, with the names of the credentials you’ll have to connect.
• How it works: whether it plans before acting, whether it delegates to subagents and what spending limit it suggests.
• Examples: two or three sample requests you can fire off with one click to see what it does.

Tip

The local alternative is not decoration. An agent that can think with Context Forge runs with no paid service behind it and no cost per use.

Offline

The gallery keeps the last catalog it downloaded. If you lose your connection you still see the extensions it already knew and their listings: it never goes blank. Once you’re back online, it refreshes on its own.

The public page

The catalog can also be browsed from a web browser, on this site’s extensions page. That is where you find each extension’s identifier, ready to paste into the gallery.

Note

The catalog’s address may change over time. You don’t need to write it down: the app discovers it on its own by reading the site’s /.well-known/contextcode.json document. That is why this documentation always links by path and never by domain.

Where next

• Want to publish your own? Head over to Build your extension.
• Rather understand the individual pieces first? Skills, MCP and plugins explains how the ones you already have installed work.