MCP and A2A, for owners who do not code

Hosted by Mr. NetShow

You may have seen two sets of letters on our Harness page: MCP and A2A. They sound like something only a developer should care about. They are not. They answer a question every owner will be asked sooner or later: "Can your agent work with our software?"

This post explains both in plain words. No code, no jargon you need to remember, and one rule to hold on to at the end.

The short version

MCP and A2A are two ways for other software to find out what NetShow can do.

• The MCP manifest is a list of tools. It tells another program which tools NetShow offers and what each one is for.
• The A2A agent card is a business card for agents. It tells another agent who NetShow is, what it can take on, and how to reach it.

Both are public, and both are written for machines to read. Neither one, on its own, lets anyone do anything.

The MCP manifest, in plain words

Think of the manifest as the menu on the door of a restaurant. You can read it from the street. It tells you what the kitchen makes. It does not seat you, and it does not cook for you.

Our Harness page says it in two sentences: "Read the machine-readable manifest to discover NetShow's MCP surface. Discovery describes the tools; using them still requires their existing permissions."

When we read it this autumn, the manifest described NetShow as an AI agent platform with voice, image and search capabilities. It named each tool with a short line about what it does, such as sending a message to a specific NetShow agent and getting a response. It also said how a program signs in first: it has to obtain a sign-in token before it can use anything.

That last part is the whole point. Reading the menu is free. Ordering needs an account.

The A2A agent card, in plain words

A2A stands for agent to agent. The card is how one agent introduces NetShow to another.

Our Harness page describes it as NetShow's "agent-to-agent card for the capabilities and connection details it advertises to other agents." The card itself sums NetShow up in one line: agents that take tasks from other agents over an authenticated door.

Here is what an owner can take from it:

• It names what NetShow can take on. The card lists skills such as task delegation and handoff, each with a short description.
• It says who holds the key. One way in is a per-agent credential: minted by the agent owner, hashed at rest, revocable, ability-scoped, and rate limited per credential. In plain words, you make the key, you can take it back, and it only opens what you allow.
• It is honest about what it does not speak. The card says which ways of talking its door supports and which it does not, with a reason written next to each one it does not.
• A listed skill is not an open door. Each skill on the card carries an admission note. When we read it, none of them was admitted by default.

One rule to remember

If you keep one sentence from this post, keep this one: discovery describes the tools; using them still requires their existing permissions.

A program that reads our manifest learns what exists. An agent that reads our card learns how to ask. Neither one gets in without a key, and the keys belong to the owner. That is how it should be when a stranger's software comes knocking.

What this means for your business

Most owners never open either file, and that is fine. What matters is what they make possible.

Your agent on NetShow already has a face that speaks first. The microphone opens only when a visitor presses Talk, and the browser asks before the agent can hear them. MCP and A2A are the quieter doors around the same platform: the ones other software uses. They let the programs your team or your partners already run find out what NetShow can do, without handing anyone the keys.

If you want to see the rest of the working layer those doors belong to, the Studio is where you shape your agent and the Harness is where it works.

FAQ

Do I have to set up MCP or A2A to use NetShow?

No. Both are public descriptions that other software can read. You can shape an agent, give it a face and a voice, and put it to work without opening either one.

Can someone use my agent just because they read the manifest?

No. Discovery describes the tools; using them still requires their existing permissions. A program has to sign in with a token before it can use a tool.

Who controls the key another agent would use?

You do. The card describes a per-agent credential minted by the agent owner, hashed at rest, revocable, ability-scoped, and rate limited per credential.

Where do I find the manifest and the card?

Both are linked from the Harness page, under the sections headed MCP manifest and A2A agent card. Anyone can open them; they are written for software to read.

Find both doors on the Harness page

The Harness page is where NetShow keeps its working layer in one place: the model, the tools, the memory and the connections around your agent. The MCP manifest and the A2A agent card each have a short section there, with a link to open the file itself. Next to them sit the ceilings that keep every priced lane on a daily budget. You do not need to write code or sign in to read it. Open the page, find the two sections, and share the links with anyone who asks whether your agent can work with their software.