Skip to main content

​
How MCP Works from a Theoretical Standpoint

Who created MCP?

Anthropic launched MCP as an open-source specification and reference server, but it is intentionally vendor-neutral so other model providers, IDE makers, and tool vendors can extend it. Anthropic seeded the common language for exposing tools, prompts, and resources to agents, then invited the ecosystem to harden and expand the protocol together so every agent can discover and call those capabilities in the same way.

Watch video

What is an MCP? What does it look like?

An MCP server is a lightweight process—usually a local CLI or a remote service—that exposes a manifest of capabilities over a websocket. The manifest lists tools (actions with parameters), resources (read-only data the agent can fetch), and prompts (starter templates). When Zencoder connects it introspects that manifest, so “what it looks like” to the agent is a structured JSON schema describing each capability plus the live stream of responses the server returns whenever the agent calls a tool.

What is the difference between Local MCP and Remote MCP?

A local MCP runs on your own machine, so the agent connects to localhost and any underlying APIs are called with the credentials stored on your laptop. That’s perfect for personal scripts but it means every teammate needs to install the MCP and keep secrets on their device. A remote MCP lives on a shared server: the server holds the credentials (e.g., AWS keys), enforces auth via tokens or SSO, and executes the tools centrally, so teammates simply connect over HTTPS and never handle the secrets directly.

Share card via email

​
Using MCP in Zencoder

What is the easiest way to add an MCP in Zencoder, from the library?

Open the three-dot menu → Tools → MCP Library, search the curated list, and pick a server that does not require credentials (like Context7). The detail pane shows the official repo plus the server command, and clicking Install adds it to your workspace instantly. Every built-in Zencoder agent (Code, Unit Test, etc.) inherits that MCP by default, so you can call its tools without configuring tokens or custom settings.

Watch video

How to add a custom MCP in Zencoder?

Go to Tools → Add custom MCP, give the integration a recognizable name, and paste the MCP server config—the same JSON/command snippet the maintainer documents (e.g., the GitHub or Context7 examples). Click Install and it joins your custom tools list. Zencoder’s built-in agents automatically inherit it, while custom Zen agents still need you to toggle it on (during creation or afterward) so you can control exactly which bespoke agents get access.

What types of MCP authentication can we find?

Local MCPs simply inherit whatever secrets you have on your machine (env vars, CLI logins, keychains), but remote MCPs put an extra door in front of the server. Some are open (no auth), others expect a static API key or bearer token, and increasingly you will see OAuth/device-code flows where the MCP shows a login link and you sign in before it issues a short-lived token. Enterprise deployments may go further with company SSO, mutual TLS certificates, or signed requests so only managed devices can connect; after that check passes, the MCP uses its own stored credentials (e.g., GitHub or AWS keys) to run the tools.

How are MCPs added differently to Zencoder agents vs. custom agents?

Every MCP you install in Tools flows automatically into the built-in Zencoder agents (Code, Unit Test, etc.) and cannot be removed there—Ask is the lone exception because it stays intentionally lightweight. Custom Zen agents, on the other hand, start with no extra MCPs; you add or remove them one by one in each agent’s tool configuration. Because the defaults inherit everything, it is worth pruning unused MCPs in Tools periodically so those core agents do not become bloated.

Share card via email

​
Jira Integration for Ticket-to-Code Workflow

How can we connect Jira to Zencoder?

Jira ships as a native integration, so no MCP is required. From any chat, open the attach-context menu (click or type @) and pick Jira; the first time, you will be prompted to connect. Click Connect to jump to the Integrations page, hit Connect next to Jira, and approve the browser flow where you choose which Jira project to authorize. Once approved, the toggle flips to Disconnect (and you can swap projects later), and the @ menu now lists your Jira issues so you can attach any ticket as context in a single click.

Watch video

What’s the benefit?

The integration removes copy/paste. Select a Jira issue from the @ menu and Zencoder injects the ticket’s title, description, and links straight into the conversation, so you can either add follow-up instructions or just send the ticket reference and let the agent run with it. Because the Integrations page exposes a quick Disconnect button plus a project selector, it is easy to revoke or re-scope access whenever priorities change.

Share card via email