How agents find you. · Learn

Once your design.md exists, your Yelmo handle is an MCP server. A teammate, a fan, or a curious agent in Claude Code installs it with one command:

claude mcp add --transport http yelmo-yourhandle https://yelmo.ai/api/mcp/yourhandle

That command does three things:

Registers yelmo-yourhandle as an MCP server in the agent's config.
Connects via HTTP transport, no local install. Public content needs no sign-in; the first time the agent reaches for anything private, you approve the connection once with your Yelmo account.
Exposes your design context, references and skills to the agent.

What the agent can do

The agent doesn't pull everything every time. It reaches for what it needs:

Design context — your design.md, tokens, and the things you refuse to do, so output matches your taste instead of generic defaults.
References — it searches your curated visual references and loads the ones it wants as images it can actually look at.
Skills — it can read a skill, or run it: an invoked skill executes on Yelmo and the agent gets back the result, not the instructions.

The agent decides. You don't push; the agent pulls. That's the whole point of MCP: context on demand. Free invocation is capped per day, so a popular skill stays fair-use rather than a runaway cost.

What this changes day to day

You stop pasting your design.md into every Claude conversation. You stop explaining "here's how I work" to teammates' agents. The agent that needs your taste fetches it. The one that doesn't, doesn't.

For a teammate: their agent now knows what you know without you being in the room.

For a hiring conversation: someone evaluating your work runs your MCP through Claude and sees, in real time, what hiring you would produce. Not a portfolio screenshot. Live output in your voice.

How to verify it works

After running the install command, ask Claude something like: "Design a settings page the way @dsaltaren would." If Yelmo is wired up, Claude reads design.md, applies the criteria, and produces work that respects your voice.

Now write yours → Get started See how it's installed → /mcp

Once your design.md exists, your Yelmo handle is an MCP server. A teammate, a fan, or a curious agent in Claude Code installs it with one command:

claude mcp add --transport http yelmo-yourhandle https://yelmo.ai/api/mcp/yourhandle

That command does three things:

Registers yelmo-yourhandle as an MCP server in the agent's config.
Connects via HTTP transport, no local install. Public content needs no sign-in; the first time the agent reaches for anything private, you approve the connection once with your Yelmo account.
Exposes your design context, references and skills to the agent.

What the agent can do

The agent doesn't pull everything every time. It reaches for what it needs:

Design context — your design.md, tokens, and the things you refuse to do, so output matches your taste instead of generic defaults.
References — it searches your curated visual references and loads the ones it wants as images it can actually look at.
Skills — it can read a skill, or run it: an invoked skill executes on Yelmo and the agent gets back the result, not the instructions.

The agent decides. You don't push; the agent pulls. That's the whole point of MCP: context on demand. Free invocation is capped per day, so a popular skill stays fair-use rather than a runaway cost.

What this changes day to day

For a teammate: their agent now knows what you know without you being in the room.

For a hiring conversation: someone evaluating your work runs your MCP through Claude and sees, in real time, what hiring you would produce. Not a portfolio screenshot. Live output in your voice.

How to verify it works

Now write yours → Get started See how it's installed → /mcp

Once your design.md exists, your Yelmo handle is an MCP server. A teammate, a fan, or a curious agent in Claude Code installs it with one command:

claude mcp add --transport http yelmo-yourhandle https://yelmo.ai/api/mcp/yourhandle

That command does three things:

Registers yelmo-yourhandle as an MCP server in the agent's config.
Connects via HTTP transport, no local install. Public content needs no sign-in; the first time the agent reaches for anything private, you approve the connection once with your Yelmo account.
Exposes your design context, references and skills to the agent.

What the agent can do

The agent doesn't pull everything every time. It reaches for what it needs:

Design context — your design.md, tokens, and the things you refuse to do, so output matches your taste instead of generic defaults.
References — it searches your curated visual references and loads the ones it wants as images it can actually look at.
Skills — it can read a skill, or run it: an invoked skill executes on Yelmo and the agent gets back the result, not the instructions.

The agent decides. You don't push; the agent pulls. That's the whole point of MCP: context on demand. Free invocation is capped per day, so a popular skill stays fair-use rather than a runaway cost.

What this changes day to day

For a teammate: their agent now knows what you know without you being in the room.

For a hiring conversation: someone evaluating your work runs your MCP through Claude and sees, in real time, what hiring you would produce. Not a portfolio screenshot. Live output in your voice.

How to verify it works

Now write yours → Get started See how it's installed → /mcp