For the complete documentation index, see /llms.txt. A single-fetch full bundle is at /llms-full.txt. Every docs page is also available as Markdown by appending .md to its URL, or by requesting it with Accept: text/markdown.

MCP Server

An MCP server that gives AI coding agents structured access to this library — the same component documentation this site renders, plus live checks against the deployed subgraph that static documentation cannot provide.

If your agent can already fetch llms.txt, the MCP server adds three things: selective retrieval (fetch one component's docs instead of a 90KB bundle), no network dependency for documentation, and two live tools that answer questions about real on-chain state.

Two ways to connect

	Hosted HTTP	Local stdio
Endpoint	/api/mcp	npx -y @p4n/erc8004-ui-mcp
Transport	Streamable HTTP	stdio
Install	none	npm
Graph API key	not used	optional
Documentation tools	yes	yes
Live subgraph tools	no	yes, with a key

The hosted endpoint is the fastest way in: no install, no key, open CORS, and a discovery manifest at /.well-known/mcp. It serves the four documentation tools only. The live tools stay local because they spend a Graph API key, and a public endpoint should not spend someone else's quota.

claude mcp add --transport http erc8004-ui https://erc8004-ui.vercel.app/api/mcp

It is a stateless dual-era server: it answers the current per-request-metadata revision (2026-07-28, including the mandatory server/discover) and the initialize handshake used by 2025-11-25 and earlier, so any current client works.

Local server

Add the server to your MCP client configuration. For Claude Code:

claude mcp add erc8004-ui --env GRAPH_API_KEY=your-graph-api-key -- npx -y @p4n/erc8004-ui-mcp

Or configure it directly:

{
  "mcpServers": {
    "erc8004-ui": {
      "command": "npx",
      "args": ["-y", "@p4n/erc8004-ui-mcp"],
      "env": { "GRAPH_API_KEY": "your-graph-api-key" }
    }
  }
}

Note

GRAPH_API_KEY is the same read-only Graph key the components use — see API Keys. It is optional: without it the four documentation tools work normally and the two live tools return setup instructions instead of results.

Documentation Tools

These read a build-time snapshot of the same registry that generates this site, so they cannot drift from what you are reading. They need no network access.

list_components

Lists every component grouped by registry. Takes an optional group filter — Providers, Identity, Reputation, Validation, or Activity.

get_component

Full documentation for one component — description, caveats, import line, usage, worked examples, an in-context composition example, the props table, and how it handles loading, error, and empty states. Accepts a name (ReputationScore) or a slug (reputation-score).

get_setup_guide

Returns any guide from this site: introduction, installation, concepts, api-keys, components, theming, or mcp. Defaults to installation.

get_types

The library's exported TypeScript definitions — the shape of the on-chain data these components render.

Live Tools

These query The Graph at runtime. They exist because the deployed subgraph schema drifts from the documented data model, and because whether a component renders anything depends on the specific agent you point it at.

check_chain_support

Introspects a chain's deployed subgraph and reports which components will actually work on it. Accepts a chain id (8453), a chain name (base), or a full agentRegistry string. Call it with no argument to list every chain the library has a subgraph for.

This catches a failure mode documentation cannot: when a deployed schema stops exposing a field that a component queries, that component breaks on that chain even though its docs are still correct. Checking first is cheaper than debugging an empty component later.

check_agent

Looks up a specific agent and reports which components will render real data for it, which will render an empty state, and which are unsupported on its chain. Use it to verify an agentRegistry / agentId pair is real before building a UI around it.

An agent with no validations will render an empty ValidationList no matter how correct your code is. This tool tells you that before you write it.

Why Both

The documentation tools answer how do I use this component. The live tools answer will it show anything. Static documentation can only answer the first — the second depends on chain state that changes independently of this library.

If you would rather not use MCP

The same documentation is published as a plain read-only JSON API, described by an OpenAPI 3.1 document. No key, no client library, open CORS:

curl https://erc8004-ui.vercel.app/api                        # endpoint index
curl https://erc8004-ui.vercel.app/api/components             # every component
curl https://erc8004-ui.vercel.app/api/components/agent-card  # one component, in full
curl https://erc8004-ui.vercel.app/api/guides/installation
curl https://erc8004-ui.vercel.app/api/chains

Add ?format=markdown to /api/components/{slug}, /api/guides/{slug} or /api/types to get the markdown rendering instead of JSON. Errors are JSON with a stable error.code, a hint saying what to do next, and an allowed list when the failure was an unknown identifier.

The specification is at /openapi.json (YAML at /openapi.yaml), and /agents.md covers when to reach for this library at all.

←Theming ERC8004Provider→