Toolkit

A catalog. Pick what fits.

The Toolkit holds reference pages for agent plugins, skill artifacts, MCP servers, and research platforms built around the Research Memex approach. Each page documents what the tool does, which hosts it runs on, and how to install it. Plugin support varies by host; a shared project name does not mean identical features everywhere.

How this catalog relates to practice

The catalog includes several kinds of relationship. They should not be mistaken for one fixed daily stack.

Kimi is a regular collaborator in the author's reasoning, writing, coding, and sustained projects, alongside work through Codex and Claude Code. Its different perspectives matter from the start of a task. The Kimi Plugin is one way to bring that collaboration into a workspace.

• Current working core: MinerU, Memex, Kimi Plugin, and Lotus Wisdom recur in active research work. Codex Plugin for Claude Code (codex-plugin-cc) (opens in a new tab) also supports review and delegation from Claude Code.
• Multi-model agent work: cc-fleet (opens in a new tab) is now the author's usual route when models need to work as agents. Vox MCP remains useful for one-shot consultations and defined workflows with prompts built into them.
• Less frequent use: Sequential Thinking remains available, but the author now uses Lotus Wisdom more often. This is a change in personal practice, not a claim that Sequential Thinking is universally incompatible with current models.
• Source infrastructure: Zotero remains useful for source and citation-library work; Zotero MCP keeps that library available to agents.
• Optional discovery: Research Rabbit remains in the teaching examples, but the author uses it less often now (October 19, 2026). It is not a required step in the research workflow.
• Onboarding: Carrel helps other researchers choose and configure tools for their work.
• Occasional experimentation: DeepThonk is used selectively and also serves as a proof-of-concept project.
• External-facing projects: Research Scanner, Interpretive Orchestration, and OpenInterviewer are built and maintained for broader use. They are not all steps in the author's routine personal workflow.

For a reasoning record that persists across sessions, see Durable Thinking. It is a separate Cloudflare deployment, not an additional step required by this catalog or something Carrel installs.

Carrel installs much of this for you

Carrel is a Claude Code plugin that interviews a researcher, audits their machine, and configures an Obsidian-centered environment. It installs the MCPs in this catalog that the interview warrants: Vox for multi-model access, MinerU for complex PDF parsing, and Zotero MCP for citation-library work. It also installs underlying conversion and transcription CLIs that do not have Toolkit pages because they are not agent-callable.

What Carrel doesn't install today:

• Other Claude Code plugins (scope, not architecture): Memex, Kimi Plugin, and Interpretive Orchestration are siblings to Carrel. The onboarding could install them and today doesn't. Install from the plugin marketplace yourself; they coexist with Carrel cleanly.
• Skill artifacts: Research Scanner is a portable skill brief, not a plugin Carrel installs.
• Methodology-specific and compute-heavy MCPs: Lotus Wisdom ships inside Interpretive Orchestration; Sequential Thinking is a separate optional server; DeepThonk is added manually when you want OpenDeepThink-style test-time compute.
• Standalone platforms: OpenInterviewer is a web app you deploy yourself.
• Persistent reasoning records: Durable Thinking needs its own Cloudflare deployment and MCP connection.

The per-tool install paths live on Carrel's own page, derived from the registry rather than restated here.

If you want the catalog without the conversation, browse the sections in the sidebar. If you want the conversation, start with Carrel and let the interview pick from this catalog on your behalf.

How the sections divide

Each tool page opens with a host support matrix: an honest reading of which hosts each tool reaches at full capability, which are adjacent (the pattern is reachable, the plugin is not), and which are out of scope.

@misc{docs-toolkit-2026,
  author = {Xule Lin},
  title = {Toolkit},
  year = {2026},
  howpublished = {\url{https://research-memex.org/docs/toolkit}},
  note = {ORCID: 0000-0001-7885-4194}
}