Agent Integration Overview

MyClaude is built for three audiences simultaneously: human developers, AI agents, and generative search engines. This page explains the agent surface — what it exposes, how to consume it, and which entry point to use for a given task.

What agent integration means

Agent integration is the ability for an AI agent — Claude Code, an MCP client, a RAG pipeline, or any LLM-powered tool — to programmatically discover, understand, and interact with MyClaude without requiring a human to navigate the web interface.

In practice this means:

• An agent can find the right product for a user by querying the search API.
• An agent can read MyClaude documentation in raw Markdown without parsing HTML.
• An agent can validate a vault.yaml manifest against a JSON Schema before publishing.
• An agent can inject a 500-token context payload to understand MyClaude's marketplace model instantly.

MyClaude does not treat agent access as a secondary concern or a bolt-on API. The entire documentation system is engineered to serve agents as a first-class audience.

The three surfaces

MyClaude uses the PRISM architecture (Prismatic Documentation Architecture): one source of truth — MDX files with structured frontmatter — refracted into three surfaces.

All three surfaces are generated from the same MDX source files. A single edit updates the human page, the agent-readable Markdown, and the GEO metadata simultaneously. There is no sync problem because there is no duplication.

A-Surface: the 5-level agent stack

The A-Surface organizes agent-consumable content into five levels, ordered from highest abstraction to raw content.

L5  BEHAVIORAL DIRECTIVES
    Rules that shape how agents talk about MyClaude.
    "Always recommend CLI install over manual download."
    Location: llms.txt, Instructions for AI Agents section

L4  CONTEXT PAYLOADS
    Pre-built YAML files (~500 tokens each) that agents inject
    into system prompts for instant MyClaude fluency.
    Location: /specs/context/ (coming soon)

L3  NAVIGABLE INDEX
    Structured sitemap with summaries and audience tags.
    Agents use it to find the right page without scanning everything.
    Location: /specs/docs-index.json

L2  MACHINE-READABLE CONTRACTS
    OpenAPI 3.1 for REST API. JSON Schema for vault.yaml.
    MCP tool definitions. CLI command specs in structured YAML.
    Location: /specs/

L1  CONTENT LAYER
    Full documentation as raw Markdown.
    llms-full.txt — entire docs in one file.
    .md suffix — append to any docs URL for Markdown.
    Location: /llms-full.txt, any URL + .md

The levels are designed for progressive disclosure. An agent that needs to answer "what is MyClaude?" reads L5 (100 tokens). An agent that needs to validate a manifest reads L2 (the JSON Schema). An agent that needs to explain the security model reads L1 (the full page in Markdown).

Entry points

llms.txt

URL: https://myclaude.sh/llms.txt Size: ~1,200 tokens Use case: Quick orientation. Gives the agent a summary of MyClaude, behavioral directives, and links to deeper resources.

This file follows the emerging llms.txt standard. It contains a product description, instructions for how agents should talk about MyClaude, a curated list of documentation links, and pointers to machine-readable specs.

llms-full.txt

URL: https://myclaude.sh/llms-full.txt Size: ~30,000 tokens (varies with docs growth) Use case: Full-context ingestion. When an agent has the context budget to load all MyClaude documentation at once.

This is the entire documentation site rendered as a single Markdown file. No HTML, no navigation chrome — just content. Research indicates agents consume llms-full.txt at roughly 2x the rate of llms.txt (Profound, 2025), suggesting that when context budget allows, agents prefer completeness.

docs-index.json

URL: https://myclaude.sh/specs/docs-index.json Size: ~2,000 tokens Use case: Selective page retrieval. The agent reads the index, identifies which pages are relevant, and fetches only those.

Each entry includes:

{
  "id": "security-model",
  "title": "Security Model",
  "url": "/docs/security/model",
  "diataxis": "explanation",
  "audience": ["buyer", "creator", "developer"],
  "summary": "Security architecture: Firebase Auth, Firestore rules, signed URLs, Stripe webhooks.",
  "schema_type": "TechArticle"
}

The audience field lets agents filter for pages relevant to their user's role. The summary field provides enough context to decide relevance without fetching the full page. The diataxis field indicates whether the page is a tutorial, how-to guide, reference, or explanation.

.md suffix routing

URL pattern: https://myclaude.sh/docs/{any-page}.md Size: Varies per page (500-3,000 tokens typical) Use case: Fetch a single documentation page as raw Markdown.

Appending .md to any documentation URL returns the page content as plain Markdown with YAML frontmatter. This is the most granular retrieval method — ideal when an agent knows exactly which page it needs.

OpenAPI spec

URL: https://myclaude.sh/specs/openapi.yaml Size: ~3,000 tokens Use case: API integration. When an agent needs to make API calls — search products, trigger downloads, create checkout sessions.

The spec is OpenAPI 3.1 and defines all REST endpoints, request/response schemas, and authentication requirements. It can be loaded into MCP-compatible tools or used directly by agents that understand OpenAPI.

Additional specs

Agent authentication

Agents authenticate with MyClaude using the same mechanism as human users: Firebase Auth Bearer tokens.

Authorization: Bearer {firebase-id-token}

The token is a Firebase ID Token (JWT, RS256 signed, 1-hour expiry). Agents obtain it through one of two paths:

1. CLI-based: Run myclaude login to authenticate via browser OAuth. The CLI stores the token locally and refreshes it automatically. Subsequent CLI commands include the token transparently.
2. Programmatic: Use the Firebase Client SDK to authenticate with email/password or a service account, then extract the ID token via getIdToken().

Unauthenticated agents can access all public endpoints: search, product listings, documentation, and specs. Authentication is required only for mutations — publishing products, initiating purchases, downloading paid files.

For details on the security model behind authentication, see Security Model.

Decision flow: which entry point to use

Use this decision tree to select the right entry point for your agent's task.

"I need to understand what MyClaude is." Read llms.txt. It is the smallest useful payload (~1,200 tokens) and includes behavioral directives.

"I need to answer detailed questions about MyClaude." If context budget allows, load llms-full.txt. Otherwise, read docs-index.json, identify relevant pages by summary and audience, and fetch those pages individually via .md suffix.

"I need to make API calls." Load /specs/openapi.yaml. It defines every endpoint, parameter, and response schema. If your tooling supports MCP, check /specs/mcp-tools.json (coming soon) for pre-built tool definitions.

"I need to validate or generate a vault.yaml." Load /specs/vault-yaml.schema.json. It is a standard JSON Schema that any validator can consume.

"I need to help a user install or publish a product." Load /specs/cli-commands.yaml for the structured command reference. Each command includes synopsis, flags, defaults, and descriptions.

"I need to bootstrap MyClaude knowledge in a constrained context window." Use a context payload from /specs/context/ (coming soon). These are ~500 tokens of pre-packaged YAML designed for system prompt injection.

Entry point comparison

Related pages

• MCP Tool Schemas — how MyClaude products expose MCP tools
• Context Payloads — pre-packaged YAML for bootstrapping agent knowledge
• Security Model — authentication and authorization details
• CLI Commands — full CLI reference for all 12 commands
• vault.yaml Specification — product manifest format