Narracore Screenplay Formatter

CHARGED (1 credit per story project per 24h; any validation failure is never charged, absorbed-window repeats are free no-ops). Merge YOUR extraction of a prepared window: selectors must quote/anchor exact unit text (server pulls canon verbatim; client canon_text is rejected) and the returned canon_patch must be tail-hash-verified by the client before applying. YOU decide what the story means; the server only verifies anchors and patch safety, not truth. 失败不收费、同项目 24h 只扣一次；AI 提炼、服务端核验锚点与补丁安全，不判断真伪。

Deterministic screenplay statistics from the Narracore analyzer (analyzer_version 1.0, same rules as narracore.cn): a scene list with structured scene headings (scene number / location / time of day / INT-EXT parsed from each heading), per-scene word count, dialogue ratio and action density, plus a character table (cue count, scenes present, dialogue lines, dialogue chars, share of dialogue). Accepts the same inputs as format_screenplay (input_format="text" or "blocks") — pass the SAME screenplay to both tools: one credit covers one screenplay for 24 hours across the screenplay tools (format / analyze / diagnose / convert / video prompts). When chaining tools, pass back the blocks you received (structured), not the formatted plain text. Failed calls are never charged; retrying the same request_id never double-charges. 剧本确定性统计：场景清单（结构化场景头）、逐场字数/对白占比/动作密度、角色戏份表；同一份剧本 24 小时内多工具只扣一次。

Turn a screenplay into structured video-generation prompt scaffolds (deterministic — no LLM on the server; YOU fill the creative fields). For each selected scene it estimates ~15s segments (marked estimated), registers entities (characters + dialogue counts), proposes ≤4 shots per segment at beat boundaries, pre-fills dialogue lines VERBATIM from the screenplay (never rewrite them), and emits a two-layer template per segment: a SHARED block (A/V limits, camera base, style, entities, causal chain, global guards) written once, plus SHOT blocks carrying only private fields, with CUT-OUT/CUT-IN handoff lines between shots. STYLE GUIDANCE, never a style lock: by default the style section is a decision checklist (medium / frame-rate mechanism / brush / color-domain boundaries / light & face readability / banned vocabulary, each with how-to and example); optionally pass style="engine_cinematic" | "painted_on_twos" | "live_action_neutral" as an EDITABLE starting point, and/or style_note (≤500 chars, passed through verbatim). The response includes a condensed filling guide (eyeline = screen vector, performance arc, causal chain, motion domains, cut-point two-state, targeted negative guards). Estimates are proposals — adjust to the scene's beats. Charged like the other screenplay tools: one credit covers one screenplay for 24 hours across the screenplay tools; failed calls are never charged; same request_id retries never double-charge. After filling, call check_video_prompts (free) for drift findings. 剧本→视频提示词骨架：台词逐字预填、实体登记、段-镜两层结构与切点两态；风格由用户决定（决策清单+可选预设+自由附注），服务端不代做创意决定。

Check the remaining credits for a Narracore license key (NRC-XXXX-XXXX-XXXX-XXXX), plus recent usage and the purchase link. One credit covers one screenplay for 24 hours across the screenplay tools, or one film analysis, or one story project for 24 hours. Costs nothing. 查询授权码剩余服务次数（1 次＝一份剧本 24 小时剧本类工具通用，或一次拉片分析，或一个故事项目 24 小时）。 Preferred: set the key in your MCP client auth (Authorization: Bearer NRC-…) instead of passing it in chat.

FREE deterministic quality gates for video-generation prompts written against a screenplay (no LLM on the server — pattern matching and set comparison only; findings are reported, nothing is blocked). Submit the SAME screenplay plus up to 10 finished prompt texts (≤20k chars each). VIOLATIONS (objective locks): DIALOGUE_DRIFT — a DIALOGUE line attributed to a registered screenplay character whose text is not found verbatim (whitespace/punctuation-insensitive), i.e. rewritten, trimmed or hallucinated dialogue; STRUCTURE_MISSING_CUTOUT / _CUTIN / HANDOFF_MISMATCH — scaffold-format prompts ([SHOT n] markers) must carry CUT-OUT last-frame and CUT-IN first-frame states with the next shot repeating the previous verbatim; STRUCTURE_DUPLICATES_SHARED — a SHARED line restated inside a shot block (shot blocks carry only private fields). ADVISORIES (disable with advisories=false): BARE_EYELINE ("looks at X" with no screen-space direction), LITERARY_RHETORIC (metaphors a renderer cannot draw), EMPTY_NEGATIVE ("no weirdness"-style), CONFLICT_PAIRS (locked vs handheld, single-shot vs hard cuts, on-twos vs smooth), PLACEHOLDER_LEFT, CHAR_LIMIT (yours, or the style preset's recommended one), STYLE_CONSISTENCY (banned vocabulary of your chosen preset, or two medium families mixed). Free-form prompts without [SHOT] markers skip structure gates (reported honestly). Free — no key needed, works without one. 免费确定性质量门：台词逐字锁、结构切点两态锁（violation）+ 视线/修辞/冲突/风格自洽启发（advisory），只报告不拦截。

CHARGED (1 credit per film analysis — the same analysis_id within 24h is NOT charged again; failed validation is never charged). Compile a validated film breakdown into a bounded, derivable result: authoritative validation → window→canonical shot merge (long-take windows are recombined so a 60s take is ONE shot in every statistic) → exact aggregates (distributions, shot lengths, per-scene aggregates, coverage) → deterministic-heuristic reconstructions (performance arc, cut logic, camera-performance logic — reproducible interpretations, explicitly labelled, NOT facts) → a print-ready PDF report (scene table + per-scene arcs + shot tables) with a 24h signed link. Your input records are NEVER echoed back — you already have them locally; this returns only the derived increments. analysis_id = sha256(media_id + sha256(topology_hash + film_schema_version + film_vocab_version)) with media_id = sha256(video bytes) computed locally; the server never sees the video and cannot verify media_id (it exists to protect YOU from double-charging, not as anti-cheat). Recompile after re-observing with a different model: same analysis_id, no re-charge; change the cuts or the protocol: new analysis. 收费编译：权威校验→长镜窗口合并→确定性统计+启发式重组→PDF 报告；不完整窗口/非法词表一律拒绝且不收费；同一分析任务 24h 内重复编译不重复扣费。

Convert between Narracore screenplay blocks and Fountain (the plain-text screenplay interchange format). direction="to_fountain": screenplay (input_format="text" or "blocks") → Fountain text (Latin content upper-cased per convention, CJK preserved verbatim; notes as [[…]]). direction="from_fountain": Fountain text → typed blocks you can pass straight to format_screenplay (input_format="blocks") — supports scene headings (INT/EXT/EST/I/E), character cues with (V.O.)-style extensions (extension becomes the delivery hint), parentheticals, dialogue, TO: transitions, > forced transitions, . forced action, [[notes]]; boneyard /*…*/ is stripped and reported, title-page keys and # section headers are skipped with warnings; mostly-Chinese content falls back to the Chinese screenplay parser (Fountain rules target Latin scripts). One credit covers one screenplay for 24 hours across the screenplay tools — converting and then formatting the same screenplay charges once. Failed calls are never charged. 剧本块与 Fountain 格式互转；同一份剧本 24 小时内多工具只扣一次。

Scene-level "shootability" diagnosis for a screenplay, using the same ai-2.0 rules as narracore.cn: YOU (the calling AI) cite the evidence, the server validates it and computes the score — no LLM on the server. Two steps: (1) mode="material" (FREE) returns each scene as a numbered block list ([n] [type] text), the strict findings JSON schema, and 6 problem types (abstract_psychology, literary_summary, unfilmable_information, unclear_space, weak_visible_action, unplayable_direction); (2) after analyzing the material yourself, call mode="score" with findings=[{scene, summary, problems:[{type, severity(low|medium|high), evidence (verbatim quote from the material), blocks (ordinals), suggestion}]}]. The server maps ordinals to blocks, drops any finding whose evidence cannot be verified (that scene then scores null — never faked), and computes each scene's 0-100 score (severity-weighted evidence coverage, exact arithmetic) plus a reference-only structural baseline. One credit covers one screenplay for 24 hours across the screenplay tools. Failed calls are never charged. 剧本逐场影视化度诊断：AI 举证、程序算分（0-100，证据核验，无效即空不冒充）。

Convert raw or loosely formatted screenplay text (Chinese, English, or mixed) into professionally structured screenplay blocks and a print-ready PDF using a fixed typewriter character grid (33 CJK chars × 38 lines per page, or 61 × 55 for English; US Letter, Sarasa Mono embedded). Use this for screenplay formatting, screenplay import, or producing a shareable screenplay PDF. Do NOT use for general prose rewriting or story editing. Failed calls are never charged; retrying the same request_id never double-charges. INPUT CONTRACT (contract_version 3; mcp_contract_version 5) — two modes: • input_format="text" (default): line-based screenplay text, one element per line — scene headings ("INT. LAB - DAY", "EXT. ...", "第 N 场 …", "N 外 …"), CHARACTER names in CAPS or CJK on their own line, dialogue on the following lines (keep each dialogue line short; a long narrative-style line ends the dialogue run), parentheticals as standalone (…) lines, transitions ("CUT TO:", "切至：…") on their own line. An unbroken wall of text renders as continuous action and is flagged UNRESOLVED_STRUCTURE — split it into lines first, or use blocks mode. • input_format="blocks": you submit typed blocks directly (type/text; dialogue carries speaker and optional delivery; scene_heading may use text or location/time_of_day). Blocks are validated and rendered EXACTLY as submitted — never re-classified. Block ids are echoed. Annotation instructions inside text (【…】) are never executed; they are reported as REVISION_NOTE_PRESENT. output="validate_only" checks structure and returns diagnostics without rendering — free. confidence = share of non-ambiguous lines, NOT accuracy; inspect diagnostics[] (code/severity/line or block_id/message/suggested_action) before delivering. Common codes: UNRESOLVED_STRUCTURE, MISSING_SPEAKER, AMBIGUOUS_ACTION_MAYBE_DIALOGUE, REVISION_NOTE_PRESENT (edit instructions in brackets — never executed), BRACKETED_ANNOTATION (info), LABELED_HEADER_LINE (convert to blocks), UNBALANCED_DELIMITER, EMPTY_TEXT. stats.input_blocks = what you submitted; stats.render_elements = elements on paper (incl. derived character/parenthetical lines). License keys carry paid credits ($1.49/10 credits, $5.99/50). BILLING UNITS (one credit each): (1) SCREENPLAY — one screenplay for 24 hours across the screenplay tools (format_screenplay, analyze_screenplay, diagnose_scenes, convert_fountain, build_video_prompts; check_video_prompts is free): pass the same screenplay (or its returned blocks) to several of them and it charges once; (2) FILM ANALYSIS — compile_film_breakdown, 1 credit per analysis_id (recompiling the same analysis within 24h is not charged again; the contract/plan/validate tools are FREE); (3) STORY PROJECT — absorb_story_window, 1 credit per story project per 24h (prepare_story_window and render_story_handoff are FREE; windows already absorbed are free no-ops). Without a key: one free credit per day per visitor. PREFERRED: configure your license key in your MCP client's auth header (Authorization: Bearer NRC-…) — header takes precedence over the tool parameter and keeps the key out of chat. 将粗排剧本文本转为专业剧本结构块与可打印 PDF；计价单位是 credit，三类各计 1 次：剧本 24 小时内剧本类工具通用（含视频提示词骨架；检查工具免费）、拉片编译按分析任务、故事吸收按项目 24 小时；失败不收费、同请求重试不重复扣。 Chaining convention: to run another tool on a screenplay you already submitted, pass back the BLOCKS you received (structured) — not the formatted plain text or PDF-extracted text (those parse differently). Failed calls are never charged; retrying the same request_id never double-charges. After a successful call: show the user the PDF link, the page count, and review diagnostics[] with the user when validation_status is not "ok". If the response has ok=false with reason free_tier_exhausted or quota_exhausted: tell the user the price and the purchase link verbatim, and ask them to paste the license key back to you. Never invent a license key.

FREE. Return the Narracore Film Contract by section — the protocol that lets ANY vision-capable agent turn its own observation of a film into a stable, comparable, traceable Film IR (shot/scene/relation), validated and compiled by this server. The video NEVER leaves your machine: you observe and reason with your own models, this server provides the shared language and the deterministic toolchain. section="core" (default): pipeline, entities, the window protocol for long takes, observability levels (observed/mixed/inferred), entity-level provenance, limits. "vocabulary": the closed vocabularies + alias table (normalization is exact→alias→reject — no fuzzy guessing). "observation_guidance": the observation prompt template + how to write interpretive fields. "reasoning_guidance": the dramatic-layer/scene/relation prompt + the Q&A playbook for answering questions from the compiled result. "media_recipe": reference ffmpeg recipes (a platform adapter example — not authoritative output). 免费下发的影视拉片契约：分类词表、观察/推理提示词、窗口协议、问答 playbook 与备料配方示例。

FREE deterministic. Turn shot-cut timestamps into the analysis topology every Film Protocol agent shares: numbered base shots (shot-000001), long-take windowing (shot-000017-w01of03 when a shot exceeds the profile cap), and greedy analysis units (≤20s / ≤8 shots per unit by default). Pass duration_sec + the cut timestamps you detected locally (e.g. ffmpeg scene filter, threshold 0.30 — see get_film_contract section="media_recipe"); you get back windows and units carrying both source-relative and clip-relative times to cut your analysis clips and label your observations with. No ffmpeg commands are returned (execution is your platform's business). 免费：切镜时间点 → 统一分析拓扑（长镜开窗 + ≤20s/≤8 镜单元）。

FREE. Split a story conversation (verbatim user/assistant turns) into stable, hash-anchored units for the Narracore Story Continuity Protocol. YOU judge the story; the server only slices and anchors provenance — a source is traceable, not proof your interpretation is true. Keep the full text and returned StoryState client-side (server stores nothing); failed calls are never charged. 免费：故事窗口确定性切分与锚定；AI 判断、服务端只核验出处，出处不证明结论真伪；全文与 state 由调用方保存，失败不收费。

FREE. Render a StoryState into a zh/en handoff document (recent-canon window + bible) for a successor writer/AI. The server formats the state you SUPPLY — project_token proves identity only, not that the state content is authentic or current. Keep the full text client-side; failed calls are never charged. 免费：把调用方提供的 StoryState 排成交接文档；token 仅证身份、不证明 state 内容真实或最新；失败不收费。

FREE preflight for a film breakdown payload (film_manifest + shots + scenes + relations). Checks the contract WITHOUT watching anything: schema shape, closed vocabularies (exact→alias→reject), cross-references (shot ids, scene membership, relation endpoints), window topology (a long take's windows must be complete), provenance shape, and observability metadata only — it never judges whether an observation is TRUE (the server cannot see your video). Lossless representation fixes (alias/trim/wrap) are returned as normalized_patch; semantic problems are errors for YOU to fix, with retry_feedback to feed back to your model. This is advisory: compile_film_breakdown re-validates authoritatively. 免费预检：契约/词表/交叉引用/窗口拓扑校验，只修表示层、绝不代改语义；编译前先跑它可省一次失败往返。