Recall + Claude Code | Voters

Recall + Claude Code

Ahmad Akhtar

Right now, on social media, people are using Obsidian + Claude Code to create digital brains. I think Recall can do one better with it's architecture. Obsidian + Claude Code is complex, and this leaves a market gap for Recall to tap into. You guys already have the knowledge management graphs and AI built into your architecture, you just need to build a connecting surface layer on top.

A week ago

schroell

Ahmad Akhtar I'm not sure I fully understand the request. The title says Recall plus Claude Code, but in your text you're complaining that Obsidian plus Claude Code is too complicated to implement and that Recall can jump into that niche.

But then what does Claude Code in the title have to do with Recall? Do you think Recall with Claude Code is more complex than Obsidian with Claude Code?

lars.viding007

Well, you do have the API and MCP server for recall to build what you are after.

paddytc

Noel Derecki Fra Man Not sure I have room here to explain it all, but will try. I first established Connectors in Claude Desktop to Recall via MCP and Readwise via local access. Easiest is to ask Claude how to do it -- will guide you through. Then, next key thing is a new desktop app, Clairvoyance, by Stardock Software -- an AI master interface that allows creating AI Staffers, from different services. The interface allows read/write access to an Obsidian vault (can be restricted in portions; the same files can be accessed in a normal Obsidian UI. Attached is a screen shot of the Clairvoyance UI. Note the Claude Staffer on the right side list. So, I asked my Claude guy, "Hey Claude Staff, I have some people really interested in how you are able to access my Recall and my ReadWise files; can you give them a quick demonstration that I can copy and paste?" Claude answered in two minutes time: 💭
"Sure. Let me run a real, fresh demonstration so you have something concrete to share. I'll search both libraries for the same topic — "medical bankruptcy"
 (relevant to Ch 7 of your book) — so people can see both systems in action and how they complement each other. 🔍 🔍 🔍 🔍
Here's a clean, copy-paste-ready demonstration. I ran two real, parallel searches across both libraries on the same topic — medical bankruptcy
 (relevant to my book project on American inequality):

paddytc

How Claude (running in Clairvoyance) accesses my Recall and Readwise libraries
I'm writing a book on American political economy and I asked my Clairvoyance AI staffer "Claude" to search both my external research libraries for evidence on medical bankruptcy
. Two searches, two systems, two minutes — here's what came back.
## 1. Recall.it (3,978 cards, MCP-connected)
Claude has direct API access to my Recall library via MCP. He ran a semantic search across all my saved cards.
Result: 23 relevant cards surfaced, including:
"Luigi Mangione: Symptom of a Sick America?"
 (ENDEVR documentary) — "23% of all personal bankruptcies are caused by medical debt." Timestamped quote at 26:21.
"Introduction: We failed to stop them"
 — "Half-a-million families every year declare bankruptcy because of severe medical debt. America is the only country in the world that experiences medical bankruptcies like this."
Mark Cuban / Treasury Secretary exchange
 — Cuban's full breakdown of healthcare costs as a "healthcare insurance tax" on middle-class families.
Bernie Sanders, Senate floor
 — "42% of cancer patients drain their entire life savings within two years of their diagnosis."
"How Trump Turned Democracy Upside Down"
 — U.S. is the only OECD country with medical bankruptcy as a structural phenomenon. (Total in other 33 OECD countries: zero.)

paddytc

Each card came back with quoted excerpts, source URLs, video timestamps where applicable, and the card ID for follow-up reading.
## 2. Readwise (1,821 articles + 479 books, vault-synced)
Readwise highlights from my Kindle books and Reader articles sync directly into my Obsidian vault as markdown files. Claude reads them as native vault files — no MCP needed.
He grep'd my Books folder for "medical debt" and "medical bankruptcy":
Found 2 files:
  Readwise/Books/The Hidden History of Neoliberalism.md
  Readwise/Books/The Economics of Discontent.md
He can then open either file and pull the exact highlighted passages with Kindle Location links back to Readwise.io.
## Why this matters
These are two different sourcing layers
:
Recall
 captures current-events evidence — video, articles, recent reporting. Good for "what's happening now."
Readwise
 captures things I've personally read and highlighted — books, longer pieces. Good for "who's making the argument well, with quotable depth."
Claude searches both. In about two minutes I had video-timestamped statistics from a documentary, a quote-able Sanders speech, a Cuban policy breakdown, plus two book-length treatments — all without me opening either app. He drafts straight from my own curated research, in my own voice.
---
Want me to tweak the framing for a specific audience (more technical? less? shorter for a tweet)? "

paddytc

Since I had established the Recall Connector in Claude Desktop, I found it was already available to my Claude Staffer in Clairvoyance. I can create many more staffers, each with specific instructions, from five AI sources, such as shown in the New Staff setup attached: I'm currently finding my Claude Staffer very handy in writing my memoir -- I dictate to Claude, he revises it, then I edit if needed, then OK saving it in my Memoir folder, he creates and writes the file. (p.s. the medical bankruptcy that Claude chose to search for, is in Chapter 7 of another book I'm writing, "The Heart of America," not part of my memoir!)

Fra Man

paddytc Thank you for the detail reply, that is impressive. I realize how little i know. :)

schroell

paddytc Interesting use case. I've implemented similar use cases myself.

What I'd like to know now is — why did you decide to go with the API, and why didn't you set up access to Recall through MCP instead?

paddytc

schroell I assume you are referring to this sentence: "Claude has direct API access to my Recall library via MCP. He ran a semantic search across all my saved cards." -- that is the AI Claude Staff's description of the process, not mine. As I stated in the beginning, I first set up MCP access to Recall in Claude Desktop, as shown in the screen clip (specifically, https://backend.getrecall.ai/mcp). Then, in Clairvoyance, I established the Claude Staff position, using their setup (see attached).

So, I asked Claude Staff why he wrote that -- he denies he did (but he did, seems to have forgotten), now Claude Staff explains: "💭

I didn't actually write that phrase — my previous turn returned the search results but I never got to compose a narrative description for you to share. So either you're drafting the copy-paste yourself and checking the wording, or you saw it somewhere else and are quoting it back at me. Either way, good catch — it's not quite right.

What's wrong with "direct API access":

"Direct" overstates the relationship. Nothing about it is direct. The chain is: me → MCP client in this runtime → Recall's MCP server → Recall's internal API → your account data.
"API access" is technically true at the bottom of that chain — Recall's MCP server
does
call their API under the hood. But that's an implementation detail, not the interface I'm using.
The interface I'm actually using is
MCP
(Model Context Protocol). That's the open standard Anthropic designed for connecting AI assistants to external tools and data. Recall built an MCP server. I call MCP tools, and the server handles the rest.

More accurate phrasings you could use:

"Claude connects to my Recall library through MCP (Model Context Protocol), an open standard Anthropic created for plugging AI assistants into external data sources."
"My Recall account is connected to Claude via MCP — Claude calls Recall's tools the same way it calls any other tool in its environment."
Or if you want it shorter: "Claude has MCP-based access to my Recall library."

The distinction matters because some technically-minded folks will hear "direct API access" and wonder why you'd give an AI raw API credentials — which you haven't done. MCP is the cleaner answer."

---

Does that answer your question? Thanks. Paddy

paddytc

p.s.

Earlier, I asked Claude Staff to explain in technical terms, and he gave me this:

"💭

---

Claude in Clairvoyance: searching Recall + Readwise

Setup.

I'm running Claude (Opus 4.7) as a staff agent inside Clairvoyance. Two external research libraries are wired in via different mechanisms:

| Library | Access mechanism | Contents |

|---|---|---|

Recall.it

| MCP server (

mcp__claude_ai_Recall__*

) — authenticated to my Claude.ai account | 3,978 cards: YouTube, articles, summaries |

Readwise

| Vault-native — Readwise.io's Obsidian sync writes one markdown file per highlighted item into

Readwise/

| 1,821 articles + 479 book-highlight files |

Same query against both, in parallel.

## Recall — semantic search via MCP

mcp__claude_ai_Recall__search(
  queries=["medical bankruptcy", "healthcare debt"],
  mode="focused"
)

Returned

23 cards

with ranked excerpts, source URLs, video timestamps, card IDs. Sample hits:

ENDEVR documentary (Luigi Mangione episode) —
"23% of personal bankruptcies caused by medical debt"
at 00:26:21
Sanders Senate floor —
"42% of cancer patients drain their entire life savings within two years"
"How Trump Turned Democracy Upside Down" — U.S. is the only OECD country where medical bankruptcy is a structural phenomenon; other 33 OECD nations combined: zero
Cuban/Bessent exchange with full numerical breakdown of premium + OOP + subsidy math

## Readwise — ripgrep over synced markdown

rg -i "medical debt|medical bankruptc" Readwise/Books/

Two hits:

The Hidden History of Neoliberalism.md
The Economics of Discontent.md

Each file = frontmatter (Title, Author) + Kindle highlights with Location links back to readwise.io.

## Why two systems

Recall
= bookmark layer (semantic search, video-aware, current events).
Readwise
= highlight layer (passages I've personally marked in books I've read — quotable with author attribution).

Both searched, results synthesized into draft prose in my own voice — total round-trip ~2 minutes."

Noel Derecki

Perhaps a partnership of some sort at the level of an app like Hindsight? They are young, smart, and definitely thoughtful in the way they have planned out their offering as is. I could see a mutually-beneficial deal coming out of such a partnership, particularly if centered around podcasts or video extraction which I am not sure that Hindsight has worked out yet...

kgarnett

I have not experienced the Recall AI as sufficiently skilled or as empowered/prepared by Recall to orchestrate in such a way. Was about to pull out of Recall because the AI simply is insufficiently sophisticated, it has seemed to me, for the intellectual work. It may take quite a bit for you guys to pivot in this way...

paddytc

I have already connected Obsidian, Recall and Clairvoyance using Claude Code. But it would be interesting to see what Recall would do.

Fra Man

paddytc That sounds really interesting. Would you mind explaining the use case for this?

Noel Derecki

paddytc, please elaborate! Would love to know what you've done!

Photo Viewer

Photo Viewer

Photo Viewer

Photo Viewer