Extension Path: For Knowledge Workers
🚀 No development background at all? Most knowledge workers can start directly with Claude.ai / Claude Desktop, without any setup. Only read
resources/setup-guide.en.mdA-D (30-45 minutes from zero) when you need to connect an MCP server (such as Gmail / Notion) or use CLI automation.
← Back to main path README · Continue here after Track A's A3 or Track B's Stage 7. Apply agentic AI to office / knowledge work.
Use Cases (Office Scenarios × How AI Helps)
The table below splits a knowledge worker's day into 7 common scenarios. Most of them are covered by Claude Desktop + MCP at Tier 1:
| Scenario | Pain point | How AI helps | Recommended tools |
|---|---|---|---|
| Email triage | 100 messages a day; priority is hard to judge | Categorize + draft replies for your review | Claude Desktop + Gmail MCP |
| Meetings → action items | You forget half of a 30-minute meeting; action items are not captured | Transcript → key decisions + action items | Otter / Zoom transcript + Claude |
| Cross-tool report aggregation | Slack / Gmail / Notion each hold part of the picture | Pull metrics + synthesize + email summary | n8n / Make / Langflow |
| Research / market intelligence | Hard to know what to ask or who to trust | Multi-source search + cross-validation + memo | Perplexity + Claude |
| Slack / messaging | Tone is hard to calibrate in sensitive situations | Rewrite + adjust tone + produce alternatives | Claude.ai |
| Notion / knowledge-base cleanup | Notes are messy, unstructured, and hard to find | Retag + classify + auto-summarize | Claude Desktop + Notion MCP |
| Documents / proposal drafts | Specs and proposals get stuck | Outline → sections → polish | Claude.ai |
💡 MCP is central for knowledge workers: new to MCP? Read Stage 5.2 — MCP Foundation. Looking for available MCP servers? See
resources/mcp-skills-catalog.en.md.
Curated Projects
💡 Want to wire your AI agent to Notion / Gmail / Outlook / Slack / Excel / Lark? Example: automatically turn Gmail messages into Notion todos. 62 commonly-used office integration tools are listed in
resources/mcp-skills-catalog.en.md(grouped by use case). The section below stays focused on workflow / integration-platform-level tools.
Workflow Tools
n8n ⭐⭐⭐⭐
Self-hostable workflow automation platform with built-in AI integration; visual node-based editor.
Best for: When you need glue between many SaaS tools (Slack + Gmail + Notion + AI).
Make.com (formerly Integromat)
Hosted workflow automation. Strong AI integration nodes.
Knowledge Worker Skills
obra/superpowers ⭐⭐⭐⭐
Brainstorming, planning, and decision-making skills.
Knowledge Management / Personal AI
khoj-ai/khoj ⭐⭐⭐⭐
| Field | Value |
|---|---|
| Stars | ★ 34k+ |
| License | AGPL-3.0 |
What it teaches: Self-hosted "second brain" — chat with web + local docs, schedule automations, build custom agents.
Best for: People wanting a self-hosted personal knowledge base + AI assistant.
Notes: AGPL-3.0 license (copyleft).
lobehub/lobe-chat ⭐⭐⭐⭐⭐
| Field | Value |
|---|---|
| Stars | ★ 76k+ |
| License | LobeHub Community License (Apache-2.0 base + commercial conditions) |
What it teaches: Deployable multi-agent chat platform — plugin marketplace, knowledge bases, team collaboration. One representative option for self-hosted AI workspaces.
Best for: Self-hosting a collaborative chat workspace.
Notes: Commercial use needs to verify the LobeHub Community License's added conditions.
langflow-ai/langflow ⭐⭐⭐⭐
| Field | Value |
|---|---|
| Stars | ★ 147k+ |
| License | MIT |
What it teaches: Visual AI-agent design platform — useful for mapping customer support, report assembly, and data-query workflows into nodes. More agent-focused than n8n (n8n is generic workflow). API / MCP server deployment is an advanced note, not something you need to learn first.
Best for: Knowledge workers who'd rather wire nodes than write Python; or anyone designing agent flows for team handoff.
Mintplex-Labs/anything-llm ⭐⭐⭐⭐⭐
| Field | Value |
|---|---|
| Stars | ★ 60k+ |
| License | MIT |
What it teaches: All-in-one private RAG workspace — upload documents, build agents, MCP-compatible, on-device by default. A self-hosted alternative to NotebookLM.
Best for: Knowledge workers wanting a NotebookLM-style tool, self-hosted, without sending data to the cloud.
MCP Servers Useful for Knowledge Workers
Communication MCP servers ⭐⭐⭐⭐
Slack / Gmail / Discord etc. The original Anthropic-hosted reference servers were reorganized in 2025; community-maintained servers now live in punkpeye/awesome-mcp-servers and wong2/awesome-mcp-servers. Browse those lists for current Slack / Gmail / Drive / Calendar MCP servers.
Workflows You Can Build (by frequency)
| Frequency | Workflow | Steps (≤3) | Recommended tools | Best for |
|---|---|---|---|---|
| Daily | Email triage | (1) Scan inbox (2) Categorize into "now / today / this week / no reply" (3) Draft replies for your review | Claude Desktop + Gmail MCP | All knowledge workers |
| Per meeting | Meetings → action items | (1) Capture transcript (Otter / Zoom) (2) Have Claude extract "key decisions + action items" (3) Assign + announce in Slack / email | Claude.ai + transcript tool | Managers / PMs |
| Weekly | Cross-tool report | (1) Pull metrics from N tools (2) Synthesize with Claude / n8n (3) Send email summary | n8n / Make / Langflow | People who send regular updates |
| Occasional | Research / market intelligence | (1) Clarify the question (2) Search multiple sources + cross-validate (3) Write a 1-2 page memo | Perplexity + Claude | Analysts / strategy roles |
| Occasional | Notion / knowledge-base cleanup | (1) Paste scattered notes into Claude (2) Ask it to retag + classify (3) Output structured Notion format | Claude Desktop + Notion MCP | Notion / Obsidian users |
💡 Starter habit: run "daily email triage" for a month first so "open inbox, open Claude" becomes natural. Adding too many automations at once is hard to sustain.
Tier Recommendations
Recommended progression:
| Tier | Tools | Best for | Learning cost |
|---|---|---|---|
| Tier 0 | Claude.ai / ChatGPT / Gemini / Perplexity (web) | Most knowledge workers start here | 0 (if you can use a browser) |
| Tier 1 | Claude Desktop + MCP (Gmail / Notion / calendar) | Repeat workflows over local / cloud files | Half a day |
| Tier 2 | n8n / Make / Langflow (automation platforms) | Connecting several SaaS tools without writing code | 1 week of setup |
| Tier 3 | Claude Code / Codex / your own Python | Dev background or dev support, production-grade workflows | Several weeks, overlaps with Track A |
Tier 3+ (CLI / SDK) is too heavy for most knowledge worker tasks. Most people can stop at Tier 1-2.
Reading
- How I Turned Claude Code Into My Personal AI Agent OS — knowledge worker case study