Recall 2.0, Six Months On: What Shipped and What Is Still Roadmap

What the board actually shows

We read Recall's public changelog at feedback.recall.it on 4 September 2026 and put every entry it renders into date order. The newest at that read was 19 August 2026 (we added the two September entries on 4 October 2026, below), the oldest still rendering is 23 May 2026, and there is no April entry on the board at all — which matters, because April is when Recall 2.0 shipped.

Recall 2.0 was 14 April 2026: the API and MCP server, the move from getrecall.ai to recall.it, the Max tier, the Product Hunt launch. That date comes from our research packet as of the packet date, 4 September 2026, not re-verified live. The board carrying no April entry today is not evidence the launch did not happen — it is most likely just how far back the rendered history goes. We are flagging it rather than quietly asserting a date we could not confirm on the page in front of us.

The timeline since April 2026

Reading the board from the bottom up, here is what Recall has shipped since the 2.0 launch, in their own headline terms.

23 May 2026 — the pop-up browser extension returns, plus multi-step AI actions.

29 May 2026 — Claude Opus 4.8 on the Max tier, a help chat that answers from the docs, in-app bug reports, a comparison hub versus NotebookLM, a subreddit, and a "Recall 2.0 form".

4 June 2026 — multi-select for export and chat, multi-card chat, better Substack saving.

15 June 2026 — group cards by date, Apple News support, passwordless code sign-in.

18 June 2026 — Custom Personas. (A 10,000-character limit on persona instructions sits in our packet against a later release; treat that number as packet-dated, not re-verified live.)

26 June 2026 — Instagram and LinkedIn saves, .txt and .md upload.

12 July 2026 — the Use Cases Hub, library filters, 250 MB PDFs.

23 July 2026 — social saves rebuilt across Instagram, TikTok, X, LinkedIn, Reddit and Facebook; table view; 62 AI languages; more models on Max. The entry describes this work as the "first step in the development of the write API".

The search release — titled 30 July 2026, dated 4 August on the board. Search moves out of the popup into the full-page library and searches reader content, notes and quizzes rather than card titles alone. There is a text-search versus AI-search toggle, results come back as real cards in grid, list or table view with multi-select and bulk actions, and search combines with tag, source, date and shared filters with matching counts.

19 August 2026 — the extension goes live on Safari (Mac) and Microsoft Edge alongside Chrome and Firefox. OCR arrives as "images as cards": upload a JPEG, PNG, WebP or HEIC and Recall reads the text, describes the picture, and generates a title, summary and tags — one image at a time. Chat gains suggested questions on cards and a panel in the desktop library. Add Content accepts up to 10 pasted links at once. Drag and drop covers PDFs, images, text files, Markdown and a Markdown zip. Unlinked mentions appear on a card's Connections tab, there is a mobile share sheet on iOS and Android, and exporting a PDF card to Markdown now includes the original PDF in the zip.

12 September 2026 — EPUB support, OCR of social-media carousels, multi-image cards and mobile search. (Added 4 October 2026.)

25 September 2026 — CSV import of up to 1,000 URLs from tools such as Raindrop and Instapaper (Plus and Max), saving chat answers to a notebook, and EPUB export. Both September entries list RSS feeds and "MCP 2.0 with write access" under Coming Next. (Added 4 October 2026.)

The 2.0 announcement was a platform. The releases since have been an application

Set the April announcement next to that list and the shape is hard to miss. Recall 2.0 announced an API and an MCP server — the two things that turn a reading tool into something other software can build on. Everything on the board since is application work: extensions on two more browsers, social saves, personas, a real search page, OCR, filters, bigger PDF limits, a share sheet.

That is a defensible order of operations, and we would rather say so than score a point. Extensions on Safari and Edge, search that reaches into reader content and quizzes, and a rebuilt social-save path across six networks are all things a paying user notices on the day they land. An API endpoint is not. The release notes read like a team deliberately making the product people actually open better.

The cost is that the platform half is the half still moving slowly, and the clearest single example is the write API.

What is still marked as coming soon

We read Recall's MCP documentation live on 4 September 2026. It says, in their own words:

"The MCP server currently provides read-only access to your knowledge base. Write capabilities are coming soon."

That page is otherwise a good piece of work. No API key is needed, authentication happens in the browser over OAuth, the requested scope is kb:read, and setup paths are documented for Cursor, Claude, ChatGPT, Perplexity and Gemini. Once connected, an assistant "can search your cards, read their content, and explore your knowledge base". What it cannot do is put anything back.

Two more things sit in the same category, on Recall's own dating. The 23 July entry calls the rebuilt social saves the "first step in the development of the write API" — a first step, named as such, in late July. The 19 August entry names bulk images and OCR on social posts as "next up". As of 4 October 2026 there is still no entry on the board announcing a shipped write API, though the 12 and 25 September entries now list "MCP 2.0 with write access" under Coming Next. We are not going to speculate about why; those are their plans, in their words, with dates attached, and that is as far as the sources go. The one structural note worth making is that the separate REST API reference at https://docs.recall.it/developer/api is read-only too — three GET endpoints — and states that a write API is on the roadmap.

Where BrainTube stands

We built in the opposite order, and it has cost us the things you would expect it to cost.

BrainTube runs one read-write MCP server covering the whole library at https://mcp.brain-tube.com/mcp, with 38 tools on every account, listed on our connect page. It connects to Claude.ai by OAuth with no API key, and MCP access and bt_ API keys are on every plan including Free. An assistant can call ingest_content or bulk_ingest — "up to 50 items in a single call with dedup and embedding" — as well as search_knowledge, and it can pull everything back out with export_corpus (a ZIP of Markdown behind a signed URL) or export_claude_md. We wrote about the read-only asymmetry in more detail in can Claude save to Recall.

What we do not have is most of the application layer above. There is no Safari or Edge extension to claim — we ship a Chrome extension and that is it. There are no native iOS or Android apps; BrainTube installs as a PWA and nothing more. We do not generate images or video. There is an Import → Recall option that takes Recall's Markdown export zip, or your own Recall API key (used once in the browser, never stored), and brings tags and source links across. We have not yet run it on a large real-world Recall library, so try it on a small export first.

Both sequencings leave a gap. Ours is the one where the tool an assistant can write into has fewer conveniences for the human sitting in front of it.

How these figures were checked

The timeline was read from Recall's public changelog at https://feedback.recall.it/changelog on 4 September 2026, which returned HTTP 200 and rendered its entries in full. The search release is titled 30 July 2026 and carries a board date label of 4 August 2026; we have not resolved which is authoritative and have not asserted a single bare date for it. The read-only quote and the OAuth and scope details come from https://docs.recall.it/developer/mcp, and the three read-only endpoints from the API reference at https://docs.recall.it/developer/api; both pages were read on 5 September 2026 and both returned HTTP 200.

Two facts here are packet-dated and were not re-verified live: the 14 April 2026 date and contents of the Recall 2.0 launch, and the 10,000-character persona instruction limit. Both are stated as of the packet date, 4 September 2026, not re-verified live. The absence of an April entry on the changelog board is not evidence against the launch date; it is most likely the limit of the board's rendered history.

Claims about BrainTube describe functionality running in production on 4 September 2026, with the tool descriptions read from https://brain-tube.com/connect.md on that date; the tool count was re-checked on 8 October 2026. If you spot something here that has gone stale, tell us and we will correct it with the date attached.