Stop your AI from inventing frameworks.

Health check: verifies the MCP is reachable and your token is valid.

Search the unified Storytelling skill corpus: method, canon, craft, examples and applied consulting/contrarian resources. The domain filter is a legacy source-family filter, not the public resource tree.

Return the full markdown for one Storytelling resource. Prefer unified URIs like `storytelling://method/agents.md`, `storytelling://canon/story-arcs.md` or `storytelling://applications/forms/activist-deck.md`; legacy ids like `storymakers/meta/agents` still work.

List logical Storytelling skill areas/categories/resources. `domain` is kept only as a legacy source-family filter.

Return curated Storymakers method markdown: overview, 10-step process, Creative Trio agents, checklists, story arcs, loop examples, examples, textures or all. This is the MCP-backed version of the Claude Storymakers skill.

Presearch the available Storymakers request lenses before asking for canon or observed slide inventory. Document kind is the primary category lens; source/source_type is preserved as producer provenance/backcompat.

Return the converged Storymakers canon: Postgres concepts.* is the source of truth for arcs, loops and tools, and MCP markdown resources are returned as companion bodies/examples. Call list_storymakers_canon_filters first when you need valid document kinds, layers, agents, slide types or component lenses.

Return the number of documents in the storytelling corpus (Postgres-backed) grouped by document kind. Documents with no classification appear under "unclassified".

Return the structured PlaybookDeck for one corpus document — slides, arc beats, loops, action titles, slide functions. Identifier accepts UUID or slug. Output is a compact prompt-friendly summary by default; pass format="json" for the full structure.

Faceted slice over corpus.page_metadata. `kind` selects the lens (slide_type, slide_function, component, chart, metric, tool, framework, arc, beat, loop, all) and `value` filters that lens. Document-level filters (kinds, sourceTypes, orientations) further narrow the scope. Returns up to `pageSize` rows (max 24).

Return the full SlideDetail for one (docId, pageNumber) — components, metrics, framework/tool/arc-beat/loop matches. Use the `docId` from search_slides results.

Meaning-first slide retrieval for deck-building agents. Describe the slide you need ("waterfall showing cost reduction across business units", "timeline of regulatory milestones with a call to act now") and get real corpus slides ranked by embedding similarity (cosine, text-embedding-3-small). Optional `document_kind` narrows to one kind (e.g. consulting-deck, pitchdeck, activist-deck); `entity_type: "deck"` searches whole decks instead of single slides. Follow up with get_slide({ docId, pageNumber }) or get_deck_playbook({ identifier }) for full detail.

Return the N best real corpus slides for one canon lens: a tool (`tool_slug`), a slide type (`slide_type_slug`) or an arc beat (`beat_slug`). Pass exactly one of the three. Exemplars are ranked by match confidence and diversified (max 2 slides per document); each carries a signed image URL you can open as a multimodal input. Optional `document_kind` narrows to one deck kind (e.g. consulting-deck).

Aggregated narrative statistics for a slice of the enriched corpus (~10k docs): arc distribution, most common beat sequence, loop usage with typical open/close pages, SCQA completeness, example big ideas, audience×intent cross-tab. Pure SQL, no LLM. Filter by document_kind (e.g. consulting-deck, pitchdeck, activist-deck), audience (e.g. investors, executives), intent (e.g. persuade-decision, raise-capital) and/or author_name substring — at least one filter is required.

Where components sit on slides of a given type, aggregated from the corpus: per component kind, the share of slides that carry it, the median bbox (x,y,w,h normalized 0–1) with p25–p75 spread, and a verbal reading ("title: top band 5–12% of height, full width"). Filter by slide_type_slug (required, e.g. kpi_overview, executive_summary, section_divider), optionally document_kind and component_kind. Pure SQL over a bounded sample, no LLM.

Audit a draft deck against corpus priors — deterministic, no LLM. Four checks: title spine (all-caps, repeated openings, outlier lengths, stat stacking), arc coverage (deck vs the modal beat sequence of the document_kind, flagging missing beats), proportions (divider/evidence/assertion shares vs the kind median), and size (slide count vs p25–p75). Pass the deck as [{page, title, slide_type?, kind_hint?}] plus document_kind (e.g. consulting-deck) and optional audience/intent to narrow the prior.

Fetch anonymized gold or flagged-bad examples for one of the 10 title species. Examples come from the 230k+ enriched corpus and are filtered against the species-specific gold criterion.

Pedagogical lookup of named title anti-patterns: definition, severity, and the canonical fix. Pass a slug for a single entry, omit it to list all 14.

Render one species ficha with live corpus data — same content as the `storytelling://copywriting/titles/species/<slug>` resource, exposed as a tool for callers that prefer the tools/call surface.

Audit a sequence of titles (in deck order) against deterministic wiring rules: all-caps share, first-word repetition, word-count outliers, stat-fact stacking, open/close echo. Returns a list of findings keyed by rule.

Diagnose a single draft title against the species taxonomy, the three quality gates (headline / forward / 8-second), the 14 anti-patterns, and the pair pattern. Backed by `claude -p` (Opus by default; ~5–25s per call).

Propose up to 3 candidate rewrites for a draft title. Each candidate must pass the three quality gates and be defensible by the evidence summary (if provided). The model declines when the evidence is too thin, returning a blocked_reason. Backed by `claude -p` (Opus default; ~10–30s per call).

⚠ Deprecated alias — will be removed 2026-07-01. Use search_storytelling_corpus({ domain: "contrarian" }) instead. Original behavior: BM25 search over the contrarian/activist application subset.

⚠ Deprecated alias — will be removed 2026-07-01. Use get_storytelling_resource({ id: "contrarian/<category>/<slug>" }) instead. Original behavior: read one contrarian/activist application markdown body.

⚠ Deprecated alias — will be removed 2026-07-01. Use list_storytelling_corpus({ domain: "contrarian" }) instead. Original behavior: list categories within the contrarian/activist application subset.

Given a thesis and audience, suggest a Storymakers arc, primary beat sequence and one supporting loop, grounded in concepts.arcs and concepts.loops.

Walk a deck draft through Storymakers checklists: action titles, narrative arc adherence, pattern coverage, slide function balance.

Bootstraps a session with the 10-step process, Creative Trio agents and quality checklists pre-loaded — for sustained authoring work.

Build a short or activist thesis using the corpus patterns: peer-gap, sum-of-parts, CEO-quote contradiction, governance, named villain.