Set up an MCP server for the Karakeep bookmark library #82

New issue

Closed

opened 2026-07-20 22:32:59 +01:00 by nimmo · 1 comment

nimmo commented

2026-07-20 22:32:59 +01:00

Owner

Goal

Expose the Karakeep (formerly
Hoarder) bookmark library as an MCP server so LLM agents can query it.

Motivation

Karakeep already holds the bookmarks I save for "look at later". The idea is to
let an agent (or a scheduled script) pull bookmarks by a specific tag and
surface them as candidates for new work — a lightweight "inbox" feeding issue
creation or direct implementation.

Proposed workflow

Stand up an MCP server backed by Karakeep's API (self-hosted instance).
Tool to list bookmarks filtered by a designated tag (e.g. to-investigate /
todo), returning title + URL + snippet.
For each surfaced bookmark:
- create a linked issue, or
- implement directly if trivial.
After a bookmark is actioned, update it in Karakeep with a new tag marking it
as seen (e.g. seen / processed) so the import loop won't pick it up
again. Alternatively, archive the bookmark once actioned.
Schedule the lookup (cron / Paseo schedule / CI) so new tagged bookmarks are
discovered without manual prompting.

Open questions

Is there a maintained Karakeep MCP server, or do we build one against the
Karakeep REST API?
Which Karakeep tag convention to standardise on, and how to avoid
double-import (tag + idempotency in the creator, not just relying on the tag)?
Should issues be filed under this milestone or a dedicated "Bookmark-driven
ideas" milestone? (Leaning toward a milestone of its own once volume grows.)

Decision criteria

Adopt if the Karakeep API exposes tag-filtered listing and tag mutation
(required for the seen/archive step), and the MCP server can run against the
self-hosted instance.

## Goal Expose the [Karakeep](https://github.com/karakeep-app/karakeep) (formerly Hoarder) bookmark library as an MCP server so LLM agents can query it. ## Motivation Karakeep already holds the bookmarks I save for "look at later". The idea is to let an agent (or a scheduled script) pull bookmarks by a specific tag and surface them as candidates for new work — a lightweight "inbox" feeding issue creation or direct implementation. ## Proposed workflow 1. Stand up an MCP server backed by Karakeep's API (self-hosted instance). 2. Tool to list bookmarks filtered by a designated tag (e.g. `to-investigate` / `todo`), returning title + URL + snippet. 3. For each surfaced bookmark: - create a linked issue, **or** - implement directly if trivial. 4. After a bookmark is actioned, update it in Karakeep with a new tag marking it as **seen** (e.g. `seen` / `processed`) so the import loop won't pick it up again. Alternatively, **archive** the bookmark once actioned. 5. Schedule the lookup (cron / Paseo schedule / CI) so new tagged bookmarks are discovered without manual prompting. ## Open questions - Is there a maintained Karakeep MCP server, or do we build one against the Karakeep REST API? - Which Karakeep tag convention to standardise on, and how to avoid double-import (tag + idempotency in the creator, not just relying on the tag)? - Should issues be filed under this milestone or a dedicated "Bookmark-driven ideas" milestone? (Leaning toward a milestone of its own once volume grows.) ## Decision criteria Adopt if the Karakeep API exposes tag-filtered listing and tag mutation (required for the seen/archive step), and the MCP server can run against the self-hosted instance.

nimmo added this to the Miscellaneous Ideas milestone

2026-07-20 22:32:59 +01:00

nimmo referenced this issue

2026-07-25 22:38:42 +01:00

feat: add Karakeep MCP integration #96

nimmo commented

2026-07-29 11:15:46 +01:00

Author

Owner

Addressed by PR #96 (feat: add Karakeep MCP integration).

nimmo closed this issue

2026-07-29 11:15:47 +01:00