Recall MCP Server

Save links, notes and tags in your Recall knowledge base.

ホスト型 MCP サーバー

npx add-mcp 'https://cotlet.dev/mcp/v1/recall'

Claude Code、Codex、Cursor などにインストールできます

ドキュメント

Cotlet · Integrations · How it works · Security · For developers · For automation agencies · Blog · API docs · Help center · Changelog · Contact · Pricing · Sign up

Recall API and MCP server: Let your AI agent fill and tidy your Recall.

Recall has had an official MCP server and REST API since April 2026, both on the Free plan. They search, list and read your cards, and nothing else: no saving, no notes, no tags, no deletes (checked 25 Oct 2026). Cotlet runs 13 Recall tools hosted, the reads plus the writes, over MCP for agents and REST for scripts.

Read Recall cards, tags and review status with an AI agent or script. Use MCP or REST to build knowledge summaries, Markdown exports and review reports.

MCP setup REST API guide

Cotlet is an independent service. It is not affiliated with, endorsed by or sponsored by Recall. Recall is a trademark of its owner. If Recall's own integrations cover what you need, use them first.

What can you automate in Recall?

  • Save a reading list as cards
  • Tag cards that have no tag
  • Write a research note from your cards
  • Send a Friday digest of new saves
  • Back up cards as Markdown
  • Clear out cards from an old site

Does Recall have an MCP server or an API?

  • Recall's MCP has 4 tools under one scope, kb:read. They cover 4 of our 13 tools in full and the tag list in part.
  • Recall lists "MCP 2.0 with write access" under "Coming Next" in its release notes of 19 Aug and 12 Sep 2026, with no date.
  • No Zapier app and no n8n node. A Zapier request on Recall's board has 75 votes since Feb 2024.
What you needOfficial MCPOfficial APICotlet
Search and read cardsYesYes, in your codeYes
Tag list and knowledge base statsYesNo endpointYes
Save links and write notesNoNoYes
Add or remove tags, delete cardsNoNoYes, deletes after your OK
Markdown export, connections, reviewNoNoYes
The route Recall providesYes, officialYes, officialUses the web app

Checked 25 Oct 2026. Sources: Recall Docs: API, Recall Docs: MCP Server, Recall changelog, Recall feature request: write API

How does Cotlet connect to Recall?

  1. Sign in to Recall. Use your own account in its Cotlet Cloud Browser.
  2. Choose MCP or REST. Connect your agent or create a Cotlet API key.
  3. Run your workflow. Ask your agent or send an HTTP request.

We never keep your Recall login, and no card is deleted until you approve it.

Connect Recall to your agent or workflow

Use MCP with your AI agent, or HTTP requests in n8n, Make, Zapier and scripts. Follow the MCP setup below or jump to the REST guide.

  1. Enable Recall for agents. Create a Cotlet account, sign in to Recall in its Cloud Browser and press Enable for agents.
  2. Add the Cotlet address. Put https://cotlet.dev/mcp/v1 into your agent as an MCP server or connector.
  3. Allow the connection. Chat apps send you to a Cotlet sign-in. Editors, n8n and scripts take an API key from the console.

The shared address, https://cotlet.dev/mcp/v1, carries Recall next to your other apps. Recall's own server can stay connected too; give it the reads and Cotlet the writes. Enabled Recall after connecting? Refresh the connector so the new tools appear.

Claude:

Settings > Connectors > Add custom connector
URL: https://cotlet.dev/mcp/v1/connect

Claude Code:
claude mcp add --transport http cotlet-recall \
  https://cotlet.dev/mcp/v1 \
  --header "Authorization: Bearer $COTLET_API_KEY"

ChatGPT:

Settings > Apps & Connectors > Create
MCP server URL: https://cotlet.dev/mcp/v1/connect
Authentication: OAuth

Cursor:

// ~/.cursor/mcp.json
{
  "mcpServers": {
    "cotlet-recall": {
      "url": "https://cotlet.dev/mcp/v1",
      "headers": { "Authorization": "Bearer ${env:COTLET_API_KEY}" }
    }
  }
}

Codex:

# ~/.codex/config.toml
[mcp_servers.cotlet-recall]
url = "https://cotlet.dev/mcp/v1"
bearer_token_env_var = "COTLET_API_KEY"

n8n:

MCP Client Tool (inside an AI Agent)
Transport: HTTP Streamable
URL: https://cotlet.dev/mcp/v1
Auth: Bearer, <your key>
Tool: recall_write

Make, Zapier:

Make: MCP Client > Call a tool
Zapier: MCP Client by Zapier > Run Tool
Server URL: https://cotlet.dev/mcp/v1
Token: <your key>
Tool: recall_read

How to use the Recall REST API

Create your Cotlet API key

Create a Cotlet API key with access to Recall and its read operations. Store it as COTLET_API_KEY in your environment or automation platform's credential store.

Sign in to Recall in its Cotlet Cloud Browser and enable the app for agents. Use your own Recall account; plan restrictions still apply to paid features.

Make your first Recall API request

cURL:

curl "https://cotlet.dev/api/v1/recall/cards/search?q=pricing&mode=focused" \
  -H "Authorization: Bearer $COTLET_API_KEY"

Read matching cards and selected text chunks

A successful REST call puts the tool output in result. This illustrative response excerpt omits call IDs, messages and credit details.

Illustrative response excerpt:

{
  "status": "success",
  "app": "recall",
  "tool": "search_cards",
  "result": {
    "documents": [
      {
        "card_id": "card_8f2a61",
        "title": "How we price a B2B SaaS",
        "created_at": "Oct 14, 2026",
        "source_url": "https://blog.example.com/how-we-price",
        "chunks": [
          {
            "chunk_id": "chunk_004",
            "content": "Start with three tiers and anchor on the middle one.",
            "source": "page 2"
          }
        ]
      }
    ],
    "total_cards": 1,
    "total_chunks": 1
  }
}

Read result.documents, where text chunks are grouped by card. Focused search selects fewer cards with more chunks; exhaustive mode selects more cards with preview chunks.

Neither search mode exposes cursor pagination. Narrow card lists by date or tag; multiple tag IDs match with OR.

get_card returns selected chunks, defaulting to 20 with a maximum of 50. Use focus_query for a topic and do not treat selected chunks as guaranteed complete card text.

Recall workflows for agents and automations

ToolConnect throughExample workflow
n8nHTTP RequestSearch cards and read selected card chunks into a research brief that retains source references.
MakeHTTP > Make a requestRead selected cards and export their Markdown into your chosen knowledge archive.
Python or Node.jsREST APICombine knowledge-base stats and review status into a personal learning dashboard.

Send Authorization: Bearer <your Cotlet API key> in request headers. You configure the schedule, destination and record comparisons.

Cotlet reads Recall; your workflow steps write to your chosen destination. These HTTP workflows do not require a native Cotlet connector or a prebuilt template.

Python and Node.js can use the same HTTP requests without an AI agent. Keep access keys in credentials, outside shared workflow exports.

The API and MCP reference lists operations and inputs. The REST API guide explains authentication, errors and approvals.

Recall API endpoints and MCP tools

Cards and search

Content comes back as text chunks with page numbers or timestamps. The whole-library zip export in Settings is not a tool; Markdown goes one card at a time.

  • Search cards (search_cards, read): Find cards by meaning or keyword, optionally within a tag, date range or source.
  • List cards (list_cards, read): Cards in your knowledge base, filtered by tag, date or source URL.
  • Get card (get_card, read): The content of one card as text chunks, with its source.
  • Get knowledge base stats (get_knowledge_base_stats, read): Total cards, date range, top tags and common sources of your knowledge base.
  • Export card as Markdown (export_card_markdown, read): The card as a Markdown file, the same as Export to Markdown in the card menu.

Saving and deleting

One link per save. On the Free plan each new card's summary counts toward Recall's 10 a month. A deleted card is gone for good.

  • Save URL (save_url, write): Save a web page, YouTube video, podcast or PDF link as a new card.
  • Create note (create_note, write): Create a new note card with a title and text.
  • Delete card (delete_card, write, asks first): Permanently delete one card.

Tags and connections

A slash nests a tag, as in Business/Pricing. Removing a tag from a card leaves the tag itself in place.

  • List tags (list_tags, read): Your tags with the number of cards under each; nested tags appear as Parent/Child.
  • Add tags to card (add_tags_to_card, write): Add one or more tags to a card; an unknown tag is created.
  • Remove tags from card (remove_tags_from_card, write): Take one or more tags off a card.
  • List card connections (list_card_connections, read): The keywords and linked cards in a card's Connections tab.

Review

Reads the Review dashboard only. It does not answer questions or start a quiz.

  • Get review status (get_review_status, read): Questions ready for review now, due this week and how many sit in each knowledge stage.

Inputs, REST path and example response of every tool: https://cotlet.dev/integrations/recall-api-mcp-server#tools

Who can see my Recall session, and how do I revoke it?

  • What we store. Your signed-in Recall profile, encrypted with AES-256-GCM, in Finland. No password or code.
  • What your agent gets. The 13 Recall tools. Never the cookie or your Recall API keys.
  • What the session sees. Your whole library: cards, notebooks, tags, review questions.
  • What no tool touches. Plan and billing, API keys, chats, personas, full export, tag deletion.
  • If Recall asks to sign in. Recall can mail a one-time code. You type it in, we never do.
  • How to cut access. Disable for agents, reset the browser, or log out of Recall there.

Our site scores A+ on Qualys SSL Labs, securityheaders.com and Mozilla Observatory (live results).

Approvals, undo and mistakes

  1. Lists the cards. Filtered by site and date, titles only.
  2. Offers a Markdown copy. Exports the cards you may want later.
  3. Waits for your OK. Each delete shows up in your Cotlet console first.
  4. Reports what went. Deleted titles, kept ones, files saved.
ActionUndo
Save a linkDelete the card; a Free summary already used stays used
Create a noteDelete the card
Add tagsRemove them with remove_tags_from_card
Remove tagsAdd them back with add_tags_to_card
Delete a cardNo, Recall has no recycle bin

1 of 13 tools asks first: delete_card. Saves, notes and tags reach nobody and undo with one call, so they run at once. You can switch the wait off for a scheduled cleanup.

A recycle bin is an open request on Recall's feedback board. Until it ships, export before you delete.

MCP call:

{
  "name": "save_url",
  "arguments": {
    "url": "https://blog.example.com/pricing-psychology"
  }
}

REST:

curl -X POST https://cotlet.dev/api/v1/recall/url/save \
  -H "Authorization: Bearer $COTLET_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url":"https://blog.example.com/pricing-psychology"}'

Tool output:

{
  "card_id": "card_e91d07",
  "title": "Pricing psychology for SaaS",
  "source_url": "https://blog.example.com/pricing-psychology",
  "status": "saved",
  "tags": [
    "Business/Pricing"
  ]
}

A cron job or script reaches the same tools over REST at /api/v1/recall, with the same key and price. It can sit next to Recall's own read API in one script: Recall for search, Cotlet for saving. The tools table lists the REST path of each tool.

How much does the Recall MCP server cost?

About $21 a month for one reader saving about 30 links a month, 3,580 credits. Your Recall plan stays with Recall.

JobCallsCredits
Save a reading list and tag it, weekly17 × 4680
Friday digest of new cards11 × 4440
Research note from saved cards7 × 4280
Markdown backup of the week, weekly11 × 4440
Tag cards that have no tag18 × 1180
Clear out cards from one old site6 × 160
Cloud Browser, 30 days1,500
Total3,580
  • Free 7-day trial, no card
  • 10 credits per successful call, MCP or REST
  • 50 credits a day for the Cloud Browser
  • Works on Recall Free, Plus ($10) and Max ($38), yearly prices

The month is our example: Recall publishes no typical library size. If your agent only reads, Recall's own MCP does it for free. On Recall Free, 8 saves a week would pass the 10 summaries a month, so Plus fits this reader. Dollars at the $29 pack rate.

Recall MCP server on GitHub, Zapier or hosted: which to use

OptionWhat it doesWhere it stops
Recall MCP serverOfficial, free, OAuth in the browser. 4 tools: search, filter, read a card, stats and tags.Read-only. Writes are announced as "MCP 2.0", no date.
Recall APIThree read endpoints with an sk_ key from Settings.No tags endpoint, no writes, code you host.
mcp-dir/recall-mcpHosted wrapper on the official API, MIT, 0 stars, last commit 21 Aug 2026.The same reads as the official server.
Zapier, n8nNo Recall app or node.Nothing to connect.
Recall extension and CSV importSaves the page you are on, or up to 1,000 links from a file.By hand; an agent cannot call them.

For search and reading, Recall's own server is free and is the route Recall backs; use it. To save, write, tag, export or delete from an agent, nothing but Cotlet does it today. Both fit in the same agent, and when Recall ships writes of its own, switch the overlap back to Recall.

Also listed on: Odel, mcpbeat

Who runs Cotlet

Take Group sp. z o.o., a software company registered in Poland, EU (VAT PL6342984273), runs Cotlet. It has worked since 2019 and was founded by Bart Krawczyk (Co-founder, CTO) and Jake Slugocki (Co-founder, SEO, sales and marketing). Send questions about the Recall tools to contact@cotlet.dev.

Recall API and MCP: questions

Can I use the Recall API without an AI agent?

Yes. Call Cotlet's REST API from a script or an HTTP step in n8n, Make or Zapier with your Cotlet key. MCP is the alternative for AI agents; you do not need to connect an agent to use REST.

How do I get access?

Sign up at https://cotlet.dev, open Recall under Integrations in the console and sign in to Recall in its Cloud Browser. The trial starts when you sign up, with no card.

How do I connect Recall to ChatGPT or Claude?

Sign up at cotlet.dev, log in to Recall once in its Cloud Browser and add Cotlet to ChatGPT or Claude, following the steps in your console. No API key is needed. ChatGPT and Claude then work in your Recall account through Cotlet, in their phone apps too.

What can ChatGPT do in Recall?

With Cotlet, ChatGPT, Claude and other AI agents handle jobs like these in Recall: save a reading list and tag it; weekly digest of what you saved; tag cards that have no tag. You ask in plain words, and anything that reaches other people waits for your OK first.

Does saving a link use my Recall AI summaries?

Yes, on the Free plan. Recall writes a concise summary into the Notebook by default, and the Free plan includes 10 AI summaries per month; Plus has unlimited summaries (Recall Pricing). You can switch to storing only the full content in your Recall settings, which Recall documents under Summarize content.

Can I get back a card my agent deleted?

No. Recall has no recycle bin, and a deleted card cannot be recovered (Recall Docs: Deleting Content). That is why delete_card asks for your OK on each call by default. Export the card as Markdown first if you may want it.

Will Recall's own MCP server get write access?

Recall says so. Its release notes of 19 Aug and 12 Sep 2026 list "MCP 2.0 with write access" under "Coming Next", without a date. On 25 Oct 2026 its docs still call the server read-only.

Does Recall have webhooks?

No. Nothing tells an agent that you saved a card. Run the agent on a schedule instead, for example a weekly list_cards from the last 7 days.

Is the MCP server different from the REST API?

Only the way in differs. Both carry the same 13 tools, inputs and answers, with one API key and one price. Agents use MCP; scripts and HTTP steps use REST.

Is Cotlet affiliated with Recall?

No. Cotlet is independent and not affiliated with Recall or recall.it.


Give your AI agents your signed-in browser.

Product: Integrations · How it works · Pricing · Security

Resources: Blog · API docs · Help center · Changelog · GitHub

Solutions: For developers · For automation agencies

Company: About · AI info: company facts · Contact

Legal: Terms of service · Privacy policy · Refund policy · Acceptable use · Cookie policy · DPA · Subprocessors

Elsewhere: Status page · Cotlet on GitHub · Cotlet on X · Cotlet on LinkedIn · Cotlet on Product Hunt · Cotlet on Crunchbase · Cotlet on SaaSHub

Cotlet.dev is built by Bart Krawczyk and Jake Slugocki, the founders of Take Group sp. z o.o., ul. Wincentego Janasa 4, 40-855 Katowice, Poland. KRS 0000871398, NIP 6342984273, VAT EU: PL6342984273. About us

Cotlet is an independent service. It is not affiliated with, endorsed by or sponsored by any of the products listed on this site. Product names are trademarks of their respective owners.