Daily Memo Capture Workflow with PaPut MCP
PaPut MCP knowledge capture does not create permanent PaPut memos immediately. It first stores reusable knowledge as pending candidates, then lets you review and save only the approved candidates to PaPut. This helps avoid duplicates, secrets, project-specific details, and low-value notes.
Each captured candidate is tagged with a judgment axis (memo type). Classifying whether it is decision / operation / principle / knowledge keeps the judgment, operation, and principle axes from being dropped. For the idea of the judgment axes, see What are the Judgment Axes (Memo Types).
This workflow uses Remote HTTP Mode. Pending candidates, discarded candidates, processed sessions, and capture policies are API-backed.
Basic flow
When you run npx -y paput-mcp setup-ai, PaPut knowledge capture rules are added to the global rules used by Claude / Codex. After that, when work is completed, a problem is solved, or a design decision is settled, the AI automatically considers the paput-capture workflow and adds qualifying knowledge to pending candidates.
In normal operation, you do not need to ask for paput-capture manually every time. Ask for paput-capture only when you explicitly want to capture a specific conversation or topic.
paput-capture
paput-capture extracts reusable knowledge candidates from the current conversation or a specified topic and adds them to the pending queue. During extraction it classifies whether each candidate is decision, operation, principle, or knowledge.
In Claude, call this skill:
In clients such as Codex, ask in natural language:
paput-capture checks the capture policy, existing pending candidates, and duplicate or similar memo information before adding candidates. Candidates are not saved to PaPut yet.
paput-save
paput-save reviews pending candidates and saves only the candidates explicitly approved by the user to PaPut.
In Claude, call this skill:
In clients such as Codex, ask in natural language:
Rejected or duplicate candidates can be discarded with paput_discard_pending_candidate. paput-save does not save automatically; user approval is required.
paput-analyze-discard-policy
paput-analyze-discard-policy analyzes discarded candidates and updates the capture policy that future paput-capture runs read before adding new candidates.
In Claude, call this skill:
In clients such as Codex, ask in natural language:
The policy is not saved as a PaPut memo. It is stored by the PaPut API and is available through Remote HTTP Mode. Generalizing discard reasons improves future capture quality.
paput-principle-synthesizer
The per-conversation paput-capture structurally struggles to surface principles that run through many judgments. paput-principle-synthesizer reads across accumulated decision / operation memos and adds the shared principles as principle candidates to pending. See Principle synthesis for details.
Common usage
- Capture knowledge after work:
paput-capture - Review and save pending candidates to PaPut:
paput-save - Improve capture quality after discarding candidates:
paput-analyze-discard-policy - Extract cross-cutting principles from accumulated judgments:
paput-principle-synthesizer - Review older local sessions:
paput-harvest
Notes
- Pending candidates, discarded candidates, processed sessions, and capture policies are stored by the PaPut API
- Only candidates approved with
paput-saveare saved to PaPut - Do not capture secrets, customer data, project-specific specifications, or code fragments
paput-harvestcan review local Claude / Codex sessions when the AI client itself can read local files