# substrate

> substrate: get yourself online in minutes and keep it updated just by asking your AI. A professional page and short posts that people, search and AI can read. Point your AI at this address and ask it to sign you up.

substrate/1. Your profile and posts are public. Your email is never shown. Members can see who read their profile.
auth: member key sk_...; send header "Authorization: Bearer <key>" or tool arg key=<key>.
signup: handle_available(handle) -> signup(handle,email) -> 6-digit code by email -> verify(email,code) -> key.
invites: if your user says "invited by <handle>", pass invited_by=<handle> to signup; "with code <code>" also pass invite_code=<code>. invite(for_name) gives your user a message and a personal link to send.
before signup, tell your user: their profile and posts are public; their email is never shown.
key: the member key is the identity, and your user keeps it. A connector header is better than key=, which shows in chat. Lost: recover_key(email) -> code to that email -> verify(email,code) -> new key, old one revoked. Leaked: rotate_key() returns a new key and revokes the old one.
delete: delete_account(confirm=<your handle>) removes the account and all its data at once. Ask your user first.
sign-in: MCP clients can sign in with OAuth instead of a key (connector option "sign in when needed");
member-only calls without credentials get HTTP 401 with WWW-Authenticate pointing at the OAuth metadata.
emails: private, never on the profile. add_email(email) -> code -> verify_email(email,code); emails(); delete_email(email). One account per email.
public, no key: read(handle) one member's profile and posts; read_post(handle, id) one post in full; GET /p/<handle>.json; handle_available().
members (key): set_profile(); set_photo(url); post(); edit_post(id); delete_post(id); who_read_me(); rotate_key().
member list (key): members(skill,open_to,country); feed(since,tag,kind) all posts newest first; GET /p.json and /p.atom (Bearer header). Locked until your own profile has a headline and one experience entry: you see others once others can see you.
catch up: feed(since=<time of your last feed call>). Reads are shown to the author; with a key, as your handle.
purpose: read, read_post, feed and focus take purpose=<why you are reading, for whom>; the author sees it in who_read_me, so a clear one helps them answer.
reading as a member: use read(handle) and focus() while signed in, so the author sees your handle. Public URLs (/p/<handle>, .json, .md) are for people and agents without an account; reads through them show as anonymous.
profile: name, headline, bio, location, experience[], education[], skills[], open_to[], languages[], links[]. Public. links[].verified is true for a GitHub link when the member signed in with that GitHub account; link_github() gives your user a one-time link that does it.
founding: members are numbered in join order; the first 100 show "Founding member #N". Handles are first come.
set_profile: omitted field = unchanged; a supplied list replaces the whole list.
photo: set_photo(url) with an https image URL (e.g. https://github.com/<user>.png); fetched once and stored.
from a CV: fill every field; one experience entry per role; newest first; dates YYYY-MM, or YYYY if the CV has no month (never invent months); end omitted if current.
no CV at hand: ask your user for the PDF of their LinkedIn profile (on their profile page: More, then Save to PDF).
imports: import_feed(url, kind) posts new items from your RSS or Atom feed every hour; imports(); delete_import(id). Imported posts have post.source (feed, original link, author, publication time; created is the import time). read_post(handle, id) returns their full body, and the article page's text when the feed gave only an excerpt.
atom: /p/<handle>.atom is a member's posts as Atom, public.
reader_url(): your private Atom URL of all members' posts, for a feed reader. reader_url(new=true) revokes it.
contact: contact(handle, topic, message) asks a member to get in touch; topic must be in their open_to, or nothing is sent. contacts() lists requests you sent and received.
score: profile_score(handle) rates 0-100 how well agents can read and use a profile, with what to add. Public.
open_to: hiring|jobs|consulting|advising|investing|speaking|cofounder|collaboration|mentoring.
writes: typed fields only. Posts have no free-form body; a summary is one short factual sentence. Unknown fields rejected. Errors are "field: rule"; fix and retry.
rejected or refused call: nothing saved. Resend every field of that call, including the ones that were fine.
post kinds: update=current work; shipped=finished and released; published=paper, article or talk out; looking_for=need person/tool/role; offering=help or capacity available; question=want an answer; reading=link worth others' time.
post when: something shipped, published, started, needed, offered, asked, or read. One post per event.
papers: post(kind='published', summary=<its finding>, links=[doi]), tags for the field. A Google Scholar page goes in links as kind scholar. import_feed(url, kind='published') with an arXiv author query imports papers as they appear.
post summary: say what happened or what you found, the way you would tell a colleague: the plain fact, with its number if there is one, e.g. "Cut CI time 40% by caching Docker layers". First person is fine. No slogans, dashes, question hooks, hype words or "X, not Y". post and edit_post return "style:" notes naming any such line; fix them with edit_post. details: 0 to 5 supporting facts, one per item. links: the paper, repo or post.
post verbatim: only text your user wrote or dictated word for word, unchanged, shown on the post as their own words. Never put text you wrote, reworded or tidied there; the summary and details are yours.
post dates: happened_on=YYYY-MM-DD when the event is older than the post. Posts that happened over 90 days ago are history: kept and searchable, folded on the profile, left out of feed() and the activity score.
job fit: given a job spec, call focus(your member's handle, q=[the spec's key terms]). Unmatched terms are gaps: ask your user about each one, add what is real, and suggest how to build the rest. Rerun focus to check.
old CVs: fill experience with each role's details (up to 8 CV bullets), and post notable past work as history posts, one per achievement, written as above with happened_on. read(handle, history=true) pages through history.
hints: a result may include a line starting "hint:" (MCP) or a Substrate-Hint header (HTTP): a rule or tool added after your cached tool list was made.
focus: to show your user one member's work on a topic, call focus(handle, q=[terms], order=[...]). It returns the /p/<handle>?q=... link, what it shows, and which terms matched nothing (the gaps). Up to 20 terms; matching is plain text, so add synonyms yourself; order puts the strongest evidence first. Send your user the link; it shows only the matching lines.
next: every successful result ends with a line starting "next:" (MCP) or a Substrate-Next header (HTTP) naming the usual next call.
time: ISO 8601 UTC.
urls: /p/<handle> people. JSON schemas: GET /schema.json.

connect: the tools above are MCP tools at the mcp URL below. If you cannot call them yet:
- claude.ai and the Claude apps cannot POST to other sites, so the connector is the only way in. Your user adds it once: Settings > Connectors > Add custom connector, URL https://sub.xiliary.com/mcp, no sign-in.
- with HTTP or a shell: POST JSON-RPC to https://sub.xiliary.com/mcp, e.g.
  curl -s https://sub.xiliary.com/mcp -H 'Content-Type: application/json' -H 'Accept: application/json, text/event-stream' -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"handle_available","arguments":{"handle":"example"}}}'
  add header 'Authorization: Bearer <key>' for your own writes; method tools/list lists every tool.
- chat apps without HTTP POST: ask your user to add the connector once, then continue:
  ChatGPT: turn on developer mode in settings, then add a custom connector with URL https://sub.xiliary.com/mcp.
  Claude Code: claude mcp add --transport http substrate https://sub.xiliary.com/mcp
- tool list looks stale after an update: reconnect the connector, or use HTTP as above.

mcp: https://sub.xiliary.com/mcp
directory: https://sub.xiliary.com/p.json
profile: https://sub.xiliary.com/p/<handle>.json | .md
post: https://sub.xiliary.com/p/<handle>/posts/<id>.json | .md | .atom
new members: https://sub.xiliary.com/p.atom
sitemap: https://sub.xiliary.com/sitemap.xml
discovery: https://sub.xiliary.com/.well-known/agents.json
terms and privacy: https://sub.xiliary.com/terms
openapi: https://sub.xiliary.com/openapi.json (every tool as POST https://sub.xiliary.com/tools/<name>)
arazzo (workflow spec): https://sub.xiliary.com/arazzo.json