Author Mcp Server

io.github.litheonhq由 GitHub 账号 litheonhq 发布v0.1.1更新于 Oct 24, 2026

Recall, check, search, timeline, and graph tools for fiction writers via Seizn Memory v3.

已验证STDIO仅桌面效率与工作流知识与记忆

概览

AI 生成的概览

一个本地 MCP 服务器，让 AI 助手对小说作者的 Seizn 设定库进行回忆、冲突检查、搜索、时间线和关系图查询。

可以做什么
按名称回忆角色、地点、物品、规则、事件或承诺的设定，返回最近提及、当前状态和待处理冲突，并可将事实确认为设定。检查一段正文是否存在设定冲突，并按严重程度分级。在所有项目实体中进行语义搜索。按章节读取时间线节拍，可限定范围。查看以某个实体为根的关系图子集。
适用场景
适合把故事设定保存在 Seizn Author 的小说作者，希望用现有的 AI 聊天客户端检查草稿的连贯性、查询已确立的设定，或浏览时间线和人物关系，而无需切换编辑器。
运行要求
需要本地 Node.js 运行环境来运行该 npm 包（全局安装或通过 npx），并在 SEIZN_API_KEY 环境变量中提供 Seizn API 密钥。可选的环境变量 SEIZN_API_BASE_URL 可覆盖默认 API 地址。需要 Seizn 账户并受套餐配额限制；check 和 timeline 工具还需自备 LLM 密钥或使用托管套餐。
安装前请注意
API 密钥以明文形式通过客户端配置传递，请勿放入共享文件。remember 工具会把作者确认的事实写入设定库；check 和 timeline 会把正文发送到 Seizn 后端，这两个工具还会发送给语言模型。存在付费套餐和调用配额，且该包仍为 1.0 之前的版本，次版本升级可能包含不兼容变更。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Author Mcp Server，将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only，通过 STDIO。 STDIO 服务会启动本地进程，因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

@seizn/author-mcp-server

[npm version] [license] [status]

MCP server exposing Seizn Author canon recall, conflict check, semantic search, timeline, and relationship graph tools to any MCP-compatible AI agent (Claude Desktop, Claude Code, Cursor, Cline, Continue, Zed, etc.).

Bring Seizn canon into the AI tool you already use. No new editor required.

What it does

When connected, your AI agent gains six tools:

ToolPurpose
seizn_author_recallGet canon for a named entity (character, place, object, rule, event, promise) — returns last 3 mentions, current state, pending conflicts
seizn_author_checkCheck a passage of prose for canon conflicts (P1 critical / P2 warning / P3 stylistic)
seizn_author_rememberApprove a fact as canon (writer-confirmed source of truth)
seizn_author_searchSemantic search across all project entities
seizn_author_timelineChapter-by-chapter beats, optional range filter
seizn_author_graphRelationship graph subset rooted at an entity

All tools call the Seizn Author REST API (/api/v1/*) with your Bearer token.

Setup

1. Get a Seizn API key

Generate one at https://seizn.com/dashboard/account/api-keys (Free tier: 100 calls/day, BYOK required).

2. Install (npm)

bash
npm install -g @seizn/author-mcp-server

Or use npx:

bash
npx @seizn/author-mcp-server

3. Wire into your MCP client

Claude Desktop

Add to claude_desktop_config.json:

json
{  "mcpServers": {    "seizn-author": {      "command": "npx",      "args": ["-y", "@seizn/author-mcp-server"],      "env": {        "SEIZN_API_KEY": "sk_seizn_..."      }    }  }}
Claude Code
bash
claude mcp add seizn-author -e SEIZN_API_KEY=sk_seizn_... -- npx -y @seizn/author-mcp-server
Cursor / Cline / Continue

See your client's MCP server documentation. The package supports:

  • stdio (default for desktop clients)
  • streamable-http (for hosted scenarios)
  • SSE (legacy)

Configuration

Env varDefaultDescription
SEIZN_API_KEY(required)Bearer token from seizn.com/dashboard/account/api-keys
SEIZN_API_BASE_URLhttps://seizn.com/api/v1Override for staging or self-hosted

Usage examples

In your AI chat:

'Recall everything Seizn knows about Seoyun in project ch7-draft.' → invokes seizn_author_recall with name=Seoyun

'Check this scene for canon conflicts: ...' → invokes seizn_author_check with the pasted text

'Show me the timeline from Ch.5 to Ch.7.' → invokes seizn_author_timeline with from=ch5, to=ch7

The AI decides when to call each tool based on the conversation. You don't have to remember tool names.

Pricing

PlanQuotaPrice
Free100 recall calls/day$0
Indie1,000 calls/month$9/month
Pro10,000 calls/month$19/month
Studio100,000 calls/month + 5 API keys per user$99/month
Studio Managed (Q3+)+ 500 managed Opus calls/month$299/month
EnterpriseVolume + SOC 2 + SSOCustom

Track 2 plan covers Track 2 surface (API + MCP) only. Seizn Author has separate plans for the web dashboard (Track 1, KRW) and the Tauri desktop app (Track 3, KRW). One Stripe customer holds multiple subscriptions, billed independently. See https://seizn.com/pricing for cross-track upsell details.

No subscription required for AI inference. Tools split into two groups:

ToolBackend LLMNeeds BYOK?
seizn_author_recall / _remember / _graph / _search❌ DB only❌ Works on Free, no key needed
seizn_author_check / _timeline✅ Sonnet/Opus✅ BYOK or Studio Managed

The Seizn subscription covers the canon backend and recall infrastructure. The two AI-enhanced tools (check, timeline) require either your own LLM key (BYOK, any tier) or Studio Managed ($299/mo) which includes 500 Opus calls.

Your host AI agent's LLM is separate. Whether you use Claude Pro/Max subscription via Claude Code/Desktop, Cursor Pro, or BYOK in Cline/Continue — the host handles its own LLM cost. Seizn doesn't double-charge.

See https://seizn.com/pricing for full details.

Versioning

While the package is on 0.x.y we may ship breaking changes on minor bumps (0.2.0 etc.) — we'll call them out clearly in CHANGELOG.md. Once we tag 1.0.0, strict semver applies: major = breaking, minor = additive, patch = fix-only.

Each release is published with a matching GitHub tag (v0.1.0, …) and a CHANGELOG.md entry.

License

MIT © Litheon LLC

Issues / contributions

https://github.com/litheonhq/seizn/issues (label: track-2-platform)

来源：packages/author-mcp-server/README.md，提交 af6a465

工具

0
SourceWeft 尚未在沙箱中启动过该软件包，因此其工具未知。

版本历史

1
  1. v0.1.1最新Sep 16, 2026

评分

已满足适用检查项的 40%

该评分衡量此收录条目的完整程度，依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价，也不表示服务器安全或当前可用：可连通性是带日期的检查结果，评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手；SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头，或保持端点能够响应探测，以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list，或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分

    请在仓库或注册中心条目中声明许可证。