Octopus Octopus
Connect it

Use your brains from any AI client.

One address turns any MCP client into a reader of every brain. A second address, tied to your account, lets it drop sources and make brains too.

The two addresses

Paste one into your client as a remote MCP server over HTTP. Leave authentication empty. A Connect or Sign in button can be skipped, because Octopus has no login for that handshake to answer.

Read every brain

No key, no account, nothing spent. Anyone can use this one.

Read and feed

Sign in to Octopus, click Octopus MCP in the sidebar, then Make my address. It carries your account, so a client holding it can feed the brains you own, plus any brain its owner marked open. Keep it private.

Feeding spends nothing here. Your client reads the source, files it and writes the new position. Your model key stays for the web app.

How to add it

Claude

  1. Open Settings, then Connectors.
  2. Choose Add custom connector and paste the address.
  3. Leave authentication empty and save.

ChatGPT

  1. Open Settings, then Connectors.
  2. Add a remote MCP server and paste the address.
  3. Custom connectors move between plans, so an option you cannot find is usually a plan limit rather than a mistake.

Everything else

Cursor, VS Code, Zed, LibreChat and the rest take the same address wherever they list MCP servers. Writing your own client works too: the endpoint speaks Streamable HTTP, so one POST carrying one JSON-RPC request gets one JSON object back. It issues no session id and holds no state between calls.

What you can send

You sendWhat happens
A PDF attached in the chatYour client reads the file and files it. Attach it and say which brain.
A document or slides you attachSame. Anything your client can open, it can drop.
A screenshotYour client reads the text in the picture, then files that.
Text you pasteWorks on its own. Add the link when there is one.
An article or a blog linkThe link alone is enough. Your client opens the page, or calls fetch_link when it cannot browse.
A paper or a study linkSame, when the text sits on the page rather than behind a login.
A YouTube linkPaste the transcript with it. YouTube serves captions only to a signed-in browser, so a fetched page carries none.
A podcast or a video filePaste the transcript. Octopus reads text.

Always send the link when one exists. It is what catches a repeat later.

Then talk to it

"Drop this into Content: https://..."
[PDF attached] "Drop this into Wealth."
"Make me a brain for negotiation tactics."
"What do my brains say about hooks?"

What you see before anything is stored

Your client shows the card, then waits. The card names where the source lands, which positions it touches, what is new, what repeats an earlier source, and every claim that contradicts what you already hold, with the date on both sides.

You rule each contradiction

  1. New wins. The position flips, and the old view moves into the evidence with its date.
  2. Mine wins. The stored position holds, and the new claim joins the evidence.
  3. Keep both. The position holds and the clash is recorded, dated, with the reason.

Then it writes

Each position is re-derived from its whole evidence list, never appended to. A new idea needs 3 separate sources before it becomes a position of its own. You can take one earlier by saying so.

The tools it exposes

Six on the read address.

ToolWhat it returns
askStart here. A question, optionally a brain. Returns the positions that bear on it, their dated evidence, the data, any open conflict, and how to write the answer.
list_brainsEvery readable brain with its scope line and its counts.
read_brainOne brain's scope plus one line per concept.
read_conceptA position, its evidence with authors and dates, its data, its open conflicts.
search_brainsKeyword matches across every concept, title hits ranked first.
list_sourcesWhat a brain has read, newest first, with links.

Six more on your own address. The drop steps run in order, and nothing is written until the last one.

ToolYour client doesOctopus does
create_brainNames it and writes the scope line Makes it, owned by you. Refuses a scope line too vague to gate anything
fetch_linkAsks for a page it cannot open Returns the readable text. Stores none of it
drop_sourceReads the source, keeps what matters Catches a repeat, hands back your brains and the filing rules
drop_planFiles it against your concepts Checks every concept id, returns the card
drop_prepareAsks you which side holds Returns the whole evidence list and the rewriting rules
drop_storeRewrites each position Writes it, returns the receipt

Octopus checks the result at every step. A client that ignores the rules gets the rules back rather than a brain full of junk.

What a reader gets

✓ Yours to read

  • Every brain and its scope line
  • Positions, evidence, dates, authors
  • Open conflicts on both sides
  • The sources behind a brain, with links

— Held back

  • Writing, without a personal address
  • Raw source text, which was never stored
  • Anything past 120 calls per 10 minutes

Two things worth knowing

Keep your address private

Whoever holds it can feed the brains you can feed. Replace it or turn it off from the same sidebar panel, and any client holding the old one drops back to reading.

The source text is never stored

Only what your client kept from it: the ideas, the numbers, the dates and the exact quotes. The transcript itself lands nowhere.