
Carrick
io.github.carrick-tools由 GitHub 帳號 carrick-tools 發布v0.4.0更新於 Nov 9, 2026
TypeScript codebase intelligence for AI agents and IDEs, across every service and repo.
概覽
遠端 MCP 伺服器,為 TypeScript 程式碼庫建立跨服務、跨儲存庫索引,讓助理依意圖搜尋函式並檢視端點、型別與呼叫端。
- 可以做什麼
- 依意圖、相似度或服務/檔案分頁瀏覽已索引的函式
- 檢視服務、專案與跨服務呼叫圖,包含未匹配的呼叫
- 查詢操作、呼叫端、API 端點及編譯器解析出的請求/回應型別
- 檢查消費端與生產端的相容性並逐項比對契約
- 讀取 npm 相依套件與版本衝突,列出對外部目標的對外呼叫
- 產生 Carrick 設定檔,例如工作流程、技能與設定腳手架
- 適用情境
- 適合在全端或多服務 TypeScript 程式碼庫中工作的助理:需要找到既有實作、在變更端點前確認誰在呼叫,或核對前端呼叫是否仍與後端處理程式一致時使用。
- 執行需求
- 遠端端點 api.carrick.tools,需透過 OAuth 登入。索引本身由 Carrick CLI 建立(npm install -g carrick、carrick init,需 Node 24 或更新版本),或在 CI 中透過 GitHub Action 更新;README 提到提供免費方案。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 Carrick,將其新增到工作區。
- 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。
- 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/io.github.carrick-tools%2Fcarrick/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。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": {
"carrick": {
"type": "http",
"url": "https://api.carrick.tools/mcp"
}
}
}README
Carrick
TypeScript codebase intelligence for AI agents & IDEs.
[Listed on mcpservers.org] [GitHub Marketplace] [Carrick MCP connector – tool definition quality and endpoint health on Glama] [Claude directory] [npm] [Open VSX] [MCP Registry]
Coding agents working across full-stack and multi-service codebases routinely rebuild existing helpers, trust stale type definitions, and modify API contracts without knowing who consumes them.
Carrick solves this by indexing your entire TypeScript ecosystem, from frontend apps to backend services, whether they live in a monorepo or across multiple repositories. By integrating deeply with the TypeScript compiler, Carrick traces every route, type, and cross-service call while recording function behaviour—so agents search by intent rather than name and stay grounded in your real architecture.
Delivered via MCP for AI agents and LSP for IDEs, Carrick ensures models see existing endpoints and utilities before generating new code. The scanner is source-available, runs from your CLI or CI pipeline, and includes a Free plan so you can start building safer agent workflows immediately.
Get started
carrick init signs you in, connects your agent and downloads your index. The quick start has the details (Node 24 or newer).
One index, three surfaces
Search existing code through your agent, inspect live contracts in your editor, and catch breaking changes in your pull requests.
In your agent
Give your AI assistants complete context on what already exists, how endpoints are shaped, and who breaks if something changes. Carrick maps every function by what it actually does, allowing agents to find existing implementations and build on them—even in repositories you haven't checked out locally. Endpoint types are pulled straight from the TypeScript compiler in the service that exposes them.
Works with any MCP-compatible agent, including Claude Code, Cursor, Windsurf, and Codex.
In your editor
Inspect cross-service contracts and jump between connected files without leaving your IDE. "Go to Definition" jumps directly from a service call to its handler when the code is on disk, while Code Lens displays every caller across your local index inline. If a frontend call and backend handler fall out of sync, Carrick surfaces type mismatches as native editor diagnostics.
Available on the Visual Studio Marketplace and Open VSX, or runnable as a standard LSP server.
In your pull request
Bring the exact same index into CI to catch contract risks, duplicate work, and silent version drift. When a producer and consumer fall out of sync—regardless of protocol—Carrick flags the exact mismatch directly on your PR before it merges. It also checks dependency versions across services to catch conflicting package updates early.
Skills
Carrick ships with four skills for the most common agent failures — breaking a caller in another service, rebuilding existing code, trusting a stale copy of another service's type, and missing places in a codebase-wide change. Each skill gets its answer from the Carrick index. carrick init installs them into .claude/skills/ and .agents/skills/.
Ask your agent
- "Which functions handle webhook signing across our services?"
- "Where do we deduplicate users by email?"
- "What calls
/api/users, and what response shape does each caller expect?" - "Show me every function that retries on rate-limit errors."
MCP tools
The MCP endpoint exposes the index as structured tools your agent can call directly. The MCP tools reference lists each tool's parameters and response.
Keep the index current
The GitHub Action refreshes each repository's part of the index on every push to main. It runs read-only in CI, with no API key to store.
Building the index has the full workflow, with on-demand rescans, dependency installs and monorepo setup.
Docs
- How Carrick works
- What Carrick covers
- carrick.json
- CLI
- MCP tools
- Task skills
- In your editor
- PR comments
Licence
Carrick has a free tier, and paid plans are on the pricing page. The scanner in this repository is source-available under the Elastic License 2.0. Copyright (c) 2026 Far Harbour B.V.
Development
See AGENTS.md for build, test, and contribution conventions.
Install the optional pre-commit hook to run formatting and tests before each commit:
來源:README.md,提交 6ae7bb8
工具
0版本歷史
1- v0.4.0最新Sep 27, 2026
評分
已滿足適用檢查項的 75%
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
請在儲存庫或註冊中心條目中宣告授權條款。