ScriptGrain

com.scriptgrain由 scriptgrain.com 發布官方發布者v1.23.1更新於 Oct 24, 2026

Your writing voice measured across 45 attributes; drafts in that voice, scored against it.

已驗證Streamable HTTP可網頁執行知識與記憶媒體與設計AI 與機器學習

概覽

AI 產生的概覽

遠端 MCP 伺服器,從 45 個面向量測你的寫作風格,用該風格起草文字,並對任何文字評分。

可以做什麼
  • 用你自己的寫作樣本或經同意的公開品牌網站建立並檢視聲音檔案
  • 依量測到的風格起草、列大綱、潤飾與去除 AI 痕跡,附風格符合分數,長文在背景執行
  • 對任何文字或網址評分、比較最多三篇文字,並給出人寫與 AI 寫的文體計量判斷
  • 管理已發表觀點與論據的記憶庫,檢查題材是否重複,並取得後續寫作角度建議
  • 用每月聲音監測追蹤已發布頁面,並維護公司資料讓草稿立基於真實事實
  • 檢視歷史草稿與版本、還原舊版本、儲存簡報範本,並查詢方案、額度與發票
適用情境
適合希望助理以你的口吻而非通用口吻寫作的場合:建立可重複使用的聲音檔案、起草或修改長文、判斷文字是否像某位作者,或避免重複自己已發表過的論點。
執行需求
遠端 streamable HTTP 端點,無需安裝。需要免費的 ScriptGrain 帳號,透過同意頁 OAuth 登入,或以 Authorization Bearer 標頭傳送 API 金鑰;也有免金鑰的訪客試用工作階段。起草、潤飾、去除 AI 痕跡與品牌模仿會消耗方案額度,部分監測與公司功能需付費方案。
安裝前請注意
起草與改寫會消耗方案額度,upgrade_plan 會回傳 Stripe 連結,但付款不會在連接器中完成。建立檔案會把你的寫作樣本傳送給 ScriptGrain,mimic_brand 在取得同意後分析第三方網站。多個工具會刪除或覆寫已存資料並標示為破壞性操作。貼上的 API 金鑰應避免納入版本控制,並在設定中撤銷不再使用的連接器金鑰。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 ScriptGrain,將其新增到工作區。
  2. 連接 所需的 憑證。它們會被私密儲存,絕不會在公開頁面上展示。
  3. 為需要使用其工具的對話啟用該服務。

Web executable,透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。

交給代理程式安裝

把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式,或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南,先向你展示發布者、端點和登入方式,經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。

閱讀 https://sourceweft.com/mcp/com.scriptgrain%2Fscriptgrain/install.md,並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前,先告訴我它的發布者、端點和登入方式,等我確認。不要把 API 金鑰寫進命令或我會提交的檔案;問我金鑰應從哪裡取得。

檢視代理程式所依據的安裝指南

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。 客戶端會在首次連接時完成 OAuth 登入。

{
  "mcpServers": {
    "scriptgrain": {
      "type": "http",
      "url": "https://mcp.scriptgrain.com/mcp"
    }
  }
}

README

ScriptGrain MCP server

ScriptGrain makes ChatGPT, Claude and the rest better at being you. This repository is the public setup guide for its remote MCP server: connect it to Claude, ChatGPT, Cursor or any MCP client, and your assistant can build a voice profile from your own writing, draft in that voice, and score any text against it.

  • Endpoint: https://mcp.scriptgrain.com/mcp (streamable HTTP, answered as plain JSON; legacy SSE at https://mcp.scriptgrain.com/sse)
  • No account yet? https://mcp.scriptgrain.com/try needs no key or sign-in: start_session gives your assistant a 7-day guest session that builds one voice profile and scores text against it, then hands you a link to keep it in a free account
  • Auth: OAuth 2.0 (sign in with your ScriptGrain account on the consent page; used by claude.ai and ChatGPT connectors), or a ScriptGrain API key sent as Authorization: Bearer sg_live_...
  • Plans: every plan, including Free. Drafting uses credits from your plan; scoring and reading are free.
  • Server version: 1.23.1, 41 tools and 2 prompts; every tool has a title and read-only / destructive hints
  • Building a profile: your assistant can use writing of yours it already holds (messages you typed to it, documents or emails you shared, your posts), quoted exactly, never its notes about you or AI-written text, and it shows you which pieces it sends
  • Website: https://scriptgrain.com · Docs: https://scriptgrain.com/docs/mcp.md · REST API: https://scriptgrain.com/docs/api.md

This repository is a guide: documentation and client configuration only. It contains no ScriptGrain source code; the application and the MCP server are private and hosted by ScriptGrain, so there is nothing to install or run.

What ScriptGrain measures

A voice profile is built from writing you have already done: 45 attributes in eight groups (lexical, syntax, tone and register, rhetoric, punctuation and format, function words, content patterns, quirks and cadence). Every draft is scored from 0 to 1 against the profile: on-voice drafts typically score 0.85 to 0.95, partial matches 0.5 to 0.8, and clearly off-voice text below 0.45. Each attribute has its own page at https://scriptgrain.com/attributes.

Connect

You need a ScriptGrain account (free): https://scriptgrain.com/auth?tab=signup&source=github_mcp

claude.ai (web) and Claude Desktop

  1. Settings, then Connectors, then Add custom connector.
  2. Paste https://mcp.scriptgrain.com/mcp.
  3. Sign in with your ScriptGrain email and password on the consent page. If you signed up with Google, open "Paste an API key instead" and paste a key from https://scriptgrain.com/settings?tab=api.

The connector mints its own API key, named after the client, in your ScriptGrain account. Revoke it in Settings, API, and the access ends.

Claude Code

With OAuth (Claude Code opens the sign-in when you run /mcp):

bash
claude mcp add --transport http scriptgrain https://mcp.scriptgrain.com/mcp

With an API key:

bash
claude mcp add scriptgrain https://mcp.scriptgrain.com/mcp \  --transport http --header "Authorization: Bearer sg_live_<your-secret>"

Project-scoped config for a shared repository: config/claude-code.mcp.json (OAuth, so no key is committed).

ChatGPT

Where your ChatGPT plan offers connectors: Settings, then Connectors, then Add custom connector, with the URL https://mcp.scriptgrain.com/mcp, and sign in with your ScriptGrain account when the consent page opens. Where connectors are not available on your plan, the free prompt builder at https://scriptgrain.com/works-with/chatgpt writes a one-page system prompt from your writing for Custom instructions instead.

Cursor (and other editors with MCP settings)

Add a remote MCP server named scriptgrain with the URL https://mcp.scriptgrain.com/mcp and complete the sign-in when the consent page opens: config/cursor.mcp.json. Where a client cannot do OAuth, send the key as a header instead: config/cursor.mcp.with-key.json. Keep a file that holds a key out of version control.

Any other MCP client

Use https://mcp.scriptgrain.com/mcp with OAuth or the Authorization: Bearer header. Every POST is answered as plain JSON, never as an event stream, so clients and proxies that only handle request and response work too. Send the Mcp-Session-Id header from the initialize response on every later call.

Tools

41 tools, listed from the server's own registrations (server 1.23.1). The tools that delete or overwrite (delete_memory, forget_piece, edit_memory, remove_monitor_url, set_voice_brain, delete_brief_template) are marked destructive, so your client can ask before running them. Credits are from your ScriptGrain plan; "free" means no credit is used.

Drafts, polish and humanise are priced by length: up to 5,000 words 1 credit, up to 10,000 words 2, up to 15,000 words 3 (the most one piece can be). A draft over 4,000 words, or a polish or humanise over 3,000 words, runs in the background: the tool returns at once with a job, and get_job shows its progress until it has succeeded (about a minute per 2,000 words). Credits are charged as the words are delivered, never above the tier asked for.

Voice profiles

ToolWhat it doesCost
list_profilesLists the voice profiles your key can use, with status, confidence score and sample countFree
create_profileBuilds a voice profile from 1 to 20 samples of your own writing (3+ pieces and 3,000+ words gives a confident profile); runs for about 1 to 3 minutes1 extraction slot
get_profileA profile's status and, once complete, its narrative description and all 45 attributesFree
mimic_brandBuilds a profile from a public brand website (up to 8 pages); asks for your explicit consent first1 brand-mimic credit

Drafting, scoring and revising

ToolWhat it doesCost
create_outlineAn outline for a long-form piece, to review before draftingFree
generate_contentA draft in your measured voice in one of 16 content types, up to 15,000 words, with its voice-match score; 1 to 3 variants (2 and 3 up to 4,000 words); over 4,000 words it is written in the backgroundPer variant: 1 credit up to 5,000 words, 2 up to 10,000, 3 up to 15,000
check_voice_matchScores any text against a profile, with the per-feature numbers behind the scoreFree
compare_voiceNo profile needed: scores up to three pieces (text or URL) against a main pieceFree
polishRevises a draft toward a voice-match target (default 0.9), up to 3 passes; over 3,000 words it runs in the background1 credit up to 5,000 words, 2 up to 10,000, 3 up to 15,000; free if the draft already passes
humanizeRewrites text (or a saved draft) to remove measurable machine-writing tells while keeping every fact, up to 3 passes; over 3,000 words it runs in the background1 credit up to 5,000 words, 2 up to 10,000, 3 up to 15,000; free if the text already passes (up to 3,000 words)
ai_detectStylometric human-vs-AI reading of any text of 50+ wordsFree
rewrite_pageRewrites a public web page's copy in your voice, block by block; prices, legal lines and button labels are left alone1 credit per 1,500 words; paid plans
save_editSaves your final edit of a draft, so the profile learns from the differenceFree
list_generationsLists past drafts with their scoresFree
get_generationFetches one past draft in full (while a long draft is being written, the part delivered so far)Free
get_jobProgress of a background job (a long draft, polish or humanise): sections done, words written, credits charged, and the result once it has succeededFree
list_draft_versionsA draft's earlier texts, newest first: every edit, polish, humanise or redraft keeps the text it replacedFree
restore_draft_versionPuts an earlier version back; the text it replaces is kept too, so it can be undoneFree
list_brief_templatesYour saved briefs (and your team's), to start from with template_id on create_outline or generate_contentFree
save_brief_templateSaves a brief (format, topic, angle, audience, register, key points, length) for next timeFree
delete_brief_templateDeletes a saved briefFree

Voice Brain: the memory of what you have already written

ToolWhat it doesCost
get_memoryThe ideas, claims, stories and frameworks your writing has covered, each with a verbatim quoteFree
sync_memoryIndexes new drafts and samples into the memoryFree
check_noveltyChecks a brief against the memory: repeats, builds on, or new, with grounded anglesFree
what_nextFive angles that build on what you keep sayingFree
add_to_memoryAdds a published piece by pasted text or URL (60+ words), or one memory stated directly (memory, with an optional kind and topic)Free
set_repeat_policyWhether drafts avoid restating earlier arguments (the default) or may repeat themFree
list_brainsLists the memories (brains) a voice may use and which voices share eachFree
set_voice_brainMoves a voice to another brain, or back to its ownFree
edit_memoryChanges one memory: its words, its kind, or its topicFree
delete_memoryDeletes one memoryFree
forget_pieceForgets a whole piece and every memory taken from itFree

Voice monitor: published pages, scored monthly

ToolWhat it doesCost
get_voice_monitorThe pages being watched, the latest run's scores and the previous runFree; paid plans
add_monitor_urlWatches a public page (Writer and Operator 5 pages, Studio 15, Agency 50)Free; paid plans
remove_monitor_urlStops watching a page; past results are keptFree
run_voice_monitorRuns this month's check now (one run per profile per month; the scheduled run is on the 1st)Free

Company context

ToolWhat it doesCost
list_companiesLists your company or product profiles, used to ground drafts in real factsFree
create_companyCreates one from a URL, freeform notes or fields; only claims the pages makeFree (reading a site by URL needs a paid plan)

Account

ToolWhat it doesCost
get_billingPlan, remaining credits and renewal dateFree
get_invoiceThe latest invoice with a PDF linkFree
upgrade_planA Stripe-hosted link to upgrade or manage the plan; nothing is charged through the connectorFree

Prompts

  • write_in_my_voice: draft a piece, report its score, iterate, and save the final edit.
  • build_my_voice_profile: gather your own writing, create the profile, and walk through the result.

Worked examples

  1. Build a voice profile from your writing and read it
  2. Draft in your voice, score it, polish it, save your edit
  3. Check whether pages sound like one writer, with no profile

Limits worth knowing

  • 60 requests per minute per key.
  • Five profile extractions an hour.
  • Drafts are written in UK or US English.
  • mimic_brand needs your explicit consent to analyse a website's writing.
  • Payments never happen through the connector.

Privacy and security

More about ScriptGrain

Support

[email protected] · https://scriptgrain.com/support

ScriptGrain is a product of Adapt Progress Evolve Limited, registered in England and Wales (10047328), London.

Licence

The documentation and configuration files in this repository are released under the MIT licence (see LICENSE). The ScriptGrain name, logo and service are not covered by it.

來源:README.md,提交 0b91d11

工具

0
該伺服器需要登入,因此 SourceWeft 無法在沒有帳號的情況下讀取其工具,也未收錄。已連接帳號所能看到的內容可能因帳號及其權限而異。

版本歷史

7
  1. v1.23.1最新Oct 15, 2026
  2. v1.21.1Oct 5, 2026
  3. v1.21.0Sep 29, 2026
  4. v1.20.0Sep 28, 2026
  5. v1.19.0Sep 28, 2026
  6. v1.18.0Sep 27, 2026
  7. v1.17.0Sep 25, 2026

評分

已滿足適用檢查項的 85%

該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通必需25 分
  • 已宣告認證方式必需10 分
  • 已列出工具15 分

    請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。

  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分