Six short, copy-paste runnable walkthroughs for the most-asked starter tasks. Each ends with explicit verification steps so you know it worked.
| # | Tutorial | Time | What you'll do |
|---|---|---|---|
| 01 | Add a new MCP tool to tapps-mcp | ~15 min | Wire a new @mcp.tool() end-to-end: handler, _record_call, checklist registration, AGENTS.md row, unit test. |
| 02 | Run the quality pipeline against a fresh Python project | ~10 min | Bootstrap with tapps_init, write a deliberately bad function, watch tapps_quick_check flag it, fix it, batch-validate, finish with the checklist. |
| 03 | Wire tapps-brain into a Claude Code session | ~20 min | Stand up the brain Docker service, set TAPPS_BRAIN_AUTH_TOKEN, save a memory in one session, recall it in the next. |
| 04 | NLT MCP session modes | ~10 min | Enable the right 1–3 NLT servers (Build / Memory / Linear / Docs) without loading all six. |
| 05 | Documentation refresh workflow | ~2 hr | Full DocsMCP-driven doc refresh: drift, links, API regen, CI gate. |
| 06 | Your first memory save and recall | ~10 min | Save and search via tapps-mcp memory CLI and nlt-memory MCP (post–v3.12.0). |
These follow the Diataxis tutorial conventions: learning-oriented, runnable end-to-end, verified at each step. They're meant for someone whose mental model of the tool is empty — not for someone who already knows what they want and needs a recipe.
For task-specific reference (the "I know what I want, just remind me how") see docs/INDEX.md. For architectural decisions, see docs/adr/.
- 02 first if you've never used the tool — it's the shortest path to seeing the pipeline work.
- 04 if you need to configure NLT MCP servers in Cursor (Build / Memory / Docs).
- 01 next if you're going to extend tapps-mcp itself (vs. just consume it).
- 03 + 06 when you want cross-session memory. Optional — tapps-mcp works without it.
- 05 when refreshing project documentation with DocsMCP.