Return the active initiative source (selected via PLAN_SOURCE env var) plus all adapters built into this server. Call first in any session so the AI knows whether to expect markdown / Linear / JIRA / Notion semantics. Returns {active, available, version}.

Enumerate product initiatives from the active source. For markdown_local this globs PLAN_PROJECT_ROOT/initiatives/*.md and reads YAML-ish frontmatter; for linear it queries the GraphQL API for issues in triage / backlog / unstarted state types; for jira it runs JQL filtered to statusCategory='To Do'; for notion it queries the database and filters to status in (Triage / Backlog / Idea). Optional filters: status (string — adapter-specific), label (string — single label match), limit (int, default 50). Returns {source, count, initiatives[]}.

Pull a single initiative by id from the active source. Returns the full Initiative record {id, source, title, body, url, status, labels, raw_metadata}. raw_metadata holds scoring inputs (reach / impact / confidence / effort / okr) plus any source-specific fields. Pair with score_initiative to get a RICE / Impact-Effort rank.

Write a new markdown_local initiative into PLAN_PROJECT_ROOT/initiatives/.md. Use this to capture an idea you (the AI client) already gathered via WebFetch / chat summary / customer-call notes — plan-master deliberately does NOT crawl URLs; you summarize, this tool persists. Only works when PLAN_SOURCE=markdown_local; for Linear / JIRA / Notion, create the issue in that platform instead. If id is omitted, auto-generates IDEA-NNN. Returns {id, written_to, source, overwritten, next_step_hint}. Typical chain: add_initiative -> score_initiative -> generate_spec_draft -> mk-spec-master.parse_spec.

Force a senior-PM analysis SOP on one initiative BEFORE scoring. Returns the initiative body + a structured checklist (target users / competition / market signal / risks / MVP scope / out-of-scope / RICE rationale) the AI client must fill in inline as its response. Loads plan-knowledge.md context if present. The tool does NOT call an LLM — it scaffolds the prompt so the AI doesn't shortcut into a shallow read. Use this WHENEVER an idea originates from chat / WebFetch and lacks a thorough product analysis. After filling the checklist, call add_initiative(overwrite=true) with the enriched body, then score_initiative. Framework options: 'default' (7 sections), 'lite' (4 sections), 'lean_canvas' (9 blocks). Returns {initiative, framework, methodology_context, analysis_checklist, instructions, next_step_hint}.

Score one initiative with RICE or Impact-Effort. Pass initiative_id to score a source-resolved record (RICE inputs are read from raw_metadata) or raw_text + overrides for an ad-hoc score without a source record. method = 'rice' (default) or 'impact_effort'. overrides = {reach, impact, confidence, effort} — any subset; takes precedence over what was in the source. RICE tier thresholds: P0 > 25, P1 10..25, P2 3..10, P3 < 3. Every call with initiative_id appends a scored decision to the index at PLAN_PROJECT_ROOT/.mk-plan-master/index.json. Returns {initiative_id, method, score, breakdown, tier, rationale, stored}.

Score every initiative the active adapter exposes and return the top-N descending. Pure arithmetic, no LLM call — the rationale string is generated from the breakdown so the output stays deterministic. Optional filters mirror list_initiatives: status, label. method defaults to 'rice'; limit defaults to 10. Auto-archives a snapshot to .mk-plan-master/history/.json so get_planning_history / get_decision_signature can compute trend deltas across cycles (debounced to 5 minutes by default). Returns {method, count, ranking[]}.

Produce a markdown spec draft for one initiative, shaped so mk-spec-master.parse_spec(raw_text=...) can ingest it verbatim. Three templates: 'default' (title / source / OKR / context / acceptance criteria / out-of-scope), 'lite' (title / context / acceptance criteria), 'detailed' (default + risks + dependencies + estimated effort). Appends a spec_generated decision to the index. Returns {markdown, suggested_filename, template_used, ready_for_mk_spec_master, next_step_hint}.

Pack the ranked backlog into a quarterly roadmap markdown, respecting an engineering capacity envelope (in engineer-months × 4 person-weeks) minus a buffer (default 20%). Uses a greedy score-per-effort packer — items with the highest RICE-per-pw ratio land first. Output is split into P0 commitments / P1 commitments / P2 stretch / Deferred / Capacity summary. Required: capacity_engineer_months (float), period (str like 'Q3 2026'). Optional: okr (str — pinned at top), method (default 'rice'), buffer_pct (default 20). Returns {markdown, scheduled[], deferred[], capacity_used_pw, capacity_total_pw, buffer_pw, method, period}.

Classify the top-N ranked initiatives into feature / tech_debt / strategic / unlabeled buckets by label, then surface ratio + score-share + a terse heuristic advisory. Use when a user asks 'is the roadmap balanced' / 'are we starving tech debt' / 'do we have any strategic bets'. Label vocabularies are configurable: feature_labels (default ['feature', 'product']), tech_debt_labels (default ['tech-debt', 'refactor', 'infra']), strategic_labels (default ['strategic', 'bet', 'moonshot']). Returns {method, totals, ratio_pct, score_share_pct, advisory}.

Create PLAN_PROJECT_ROOT/plan-knowledge.md from a starter template. The file carries methodology (RICE, WSJF, Impact-Effort, OKR mapping, INVEST, personas / job-stories, decision-log convention) plus TODO sections for active OKRs / personas / strategic bets / tech-debt zones / glossary / roadmap rhythm. Other mk-plan-master tools lean on this indirectly via get_plan_context. Idempotent — refuses to overwrite an existing file unless overwrite=true. Optional project_name labels the file. Override location via the PLAN_KNOWLEDGE_FILE env var.

Read PLAN_PROJECT_ROOT/plan-knowledge.md (or fall back to built-in defaults if missing). Call near the start of a planning session so the same methodology + domain glossary colours every scoring decision that follows. Optional section filters to a single heading (partial-match, case-insensitive) — e.g. section='RICE' returns just the RICE block. Returns {source: 'file'|'builtin', content, ...}.

Return trend deltas (current vs ~7 days ago / vs window_days ago) for the top-10 RICE-ranked backlog snapshots archived by rank_backlog. Surfaces churn (entries added/dropped) plus the average score of the current top-10. Use when a user asks 'are we improving' / 'show me the trend' / 'is the same idea always at the top'. Returns {snapshots_count, trend_7d, trend_30d, summary}.

Scan history + index for chronic patterns: ghost initiatives (appear in top-10 in >50% of snapshots but never spec_generated), score whiplash (RICE swings >50% between snapshots → bad data quality), orphan OKRs (OKRs in the index with zero initiatives in the current top-10). Use when a user asks 'which ideas keep getting punted' / 'why does this score keep moving' / 'which OKR has no execution'. Args: window_days (default 30). Returns {ghost_initiatives, score_whiplash, orphan_okrs, summary}.

Aggregate the tool-usage log written by this server. Surfaces: which tools are called most, which fail most (error rate), p50 / p95 / p99 latency, and which declared tools have never been called in the window (dead surface). Records contain only tool name + timing + ok flag — argument values are never logged. Use when a user asks 'what's the AI actually using' / 'which tools are slow' / 'which tools are unused'. Args: window_days (default 7). Returns {calls_total, calls_by_tool, error_rate_pct, p50_ms, p95_ms, p99_ms, top_tools[], dead_tools[]}.