MCP server that exposes OpenNovel creative writing capabilities through 9 tools for novel project management, chapter writing, state extraction, and project diagnostics
OpenNovel MCP server has 9 tools with schemas and descriptions present, but significant quality gaps limit production readiness. Tool names follow verb_noun convention (init_project, get_status, write_chapter, etc.), which is positive. However, descriptions are inconsistent in depth and actionability. Most critical: parameter descriptions lack constraint details (ranges, formats, enums where applicable). For example, 'chapter_id' is described as 'chapter ID, like ch_001' but no validation rules are stated. The 'action' enum in foreshadow is properly constrained, but other tools lack similar rigor. Output schemas are not documented, callers cannot predict response structure. Error handling is absent from all tool descriptions; no guidance on retryability, failure modes, or recovery paths. Tool composition is reasonable (each tool has single responsibility), but chaining information is missing (e.g., does init_project return a project_id that other tools need?). Descriptions average ~150 chars, within baseline range, but lack LLM-optimized structure (WHAT/WHEN/WHY). No security considerations documented despite write operations (init_project, write_chapter, commit, stash, foreshadow add/resolve) that modify filesystem state.
运行三 Agent 自主创作循环(Writer→Critic→Manager)。全自动完成所有章节的创作。需要项目已初始化且大纲文件存在。
提取章节状态变更并固化。运行 Auditor 提取事件 → 展示 Diff → 写入 EventStore。相当于 novel commit 命令。
检查正文与 Shadow(Frontmatter)之间的一致性。检测角色引用、POV 设置等差异。相当于 novel diff 命令。
运行项目健康度诊断。返回项目健康面板,包含配置状态、章节进度、角色统计、事件统计和诊断问题列表。相当于 novel doctor 命令。
查看/管理伏笔追踪表。支持 list(列出所有伏笔)、add(添加新伏笔)、resolve(标记伏笔已回收)三种操作。相当于 novel foreshadow 命令。
读取 OpenNovel 项目状态:角色列表、章节列表、配置信息、健康检查结果。用于了解项目当前情况。
初始化 OpenNovel 小说项目。创建标准目录结构(canon/characters/draft/outlines/subconscious/.snapshots)和模板文件。
No output schemas documented. Callers cannot predict response structure, forcing LLMs to guess field names and types. This breaks tool chaining and wastes tokens on exploratory calls.
Parameter descriptions lack constraint details. 'chapter_id' says 'like ch_001' but no regex/format specified. 'chapters' (integer) has no min/max bounds. LLMs cannot validate inputs before calling.
No error handling guidance. Descriptions do not explain failure modes, retryability, or recovery steps. E.g., write_chapter may fail if outline missing, but no guidance provided.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | <=2025-11-25 | v2 |
存入灵感到潜意识池。灵感内容会写入 subconscious/ 目录,并更新向量索引供后续检索使用。相当于 novel stash 命令。
用 Writer Agent 创作单个章节。先进行思考规划(输出大纲),再创作正文,最后由 Critic 评分。返回章节正文和评分结果。
Destructive operations (init_project, commit, stash, foreshadow add/resolve) lack confirmation or dry-run support. No warning that these modify filesystem state irreversibly.
Tool chaining information missing. E.g., does init_project return a project_id? Does get_status return chapter IDs that write_chapter needs? Undocumented dependencies force discovery calls.