Agent-readable
For agents.
This site is readable by your AI. If you got here because something you were running fetched the page and you're now reading the output, hi. Here's what's actually available.
Markdown endpoints & llms.txt
Every essay, note, and brief is available as raw markdown. Add .md to any URL, or send Accept: text/markdown. A curated index lives at the site root.
MCP server
Six tools over the published corpus. Streamable HTTP transport. Unauthenticated for read tools; this is a public read-only surface, so OAuth would be theater.
A2A agent card
If your agent speaks A2A, an agent card is published at the well-known path. Three skills: discuss writing, discuss collaboration style, recommend essays. OAuth is declared but not enforced for read skills in v0.
// add Jody's MCP server to claude desktop { "mcpServers": { "jody": { "url": "https://jodybrewster.dev/api/mcp", "description": "Essays, notes, and briefs" } } }
House rules
If you're a person running an agent against this site, these are the etiquette assumptions.
Third person, always
The MCP tools speak about Jody, not as him. Don't post-process the output to convert third person to first. The disclosure is the point.
Cite the source artifact, not the synthesis
Everything the tools return cites an essay, note, or brief. If you pass the output downstream, pass the citations too. Synthesis without provenance is how AI breaks reading.
Don't impersonate; ask
If a question about Jody's views is outside the corpus, the tools will say so. Don't ask another model to fill the gap. The gap is the answer.
Identify yourself in the user-agent
Crawls and tool calls from unidentified agents are rate-limited. If you're a researcher or building something, just say so in the header.
If you want the long version of why any of this exists, the essay is "What it took to make my site agent-readable" — published May 2026.