
Dataiku MCP
io.github.clssck由 GitHub 帳號 clssck 發布v1.1.1更新於 Oct 27, 2026
MCP server for Dataiku DSS project, flow, and operations APIs.
概覽
本機 MCP 伺服器,讓助理透過 Dataiku DSS REST API 檢視並操作專案、流程、資料集、配方、作業、情境與資料夾。
- 可以做什麼
- 列出、讀取並產生專案流程地圖,包含配方子類型與連線關係
- 列出、檢視、預覽、下載、建立、更新與刪除資料集與配方
- 列出、檢視、建置、等待與中止作業,並讀取作業日誌
- 列出、執行與管理情境,讀取或設定專案變數
- 瀏覽受管資料夾及其內容,上傳、下載與刪除檔案
- 推斷連線資訊,列出程式碼環境
- 適用情境
- 適合在 Dataiku DSS 中工作、希望助理梳理專案流程、查看資料集結構與預覽、執行或監控建置與情境、管理資料夾與變數而不必切換到 DSS 介面的情況。
- 執行需求
- 以 Node.js 20+ 的 npm 套件(dataiku-mcp)透過 stdio 在本機執行。需要 DATAIKU_URL 提供 DSS 實例網址、DATAIKU_API_KEY 提供 DSS API 金鑰;DATAIKU_PROJECT_KEY 為選用的預設專案。需能連線至該 DSS 實例。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 Dataiku MCP,將其新增到工作區。
- 為需要使用其工具的對話啟用該服務。
Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。
其他 MCP 客戶端
參照 儲存庫 中的啟動說明。
README
[!IMPORTANT]
Sunset — use
clssck/dataiku-sdkinsteadDataiku MCP Server is no longer actively maintained. Development has moved to the Dataiku DSS SDK — a schema-first TypeScript SDK and
dssCLI that is more actively maintained and covers far more of the DSS API:
- Broader coverage — 30+ DSS resources vs this server's narrow tool set.
- A scriptable
dssCLI with a machine-readable command contract for agents.- One-command agent skill install for Claude, Codex, Cursor, Pi, and OMP.
Migrate here → https://github.com/clssck/dataiku-sdk
Dataiku MCP Server
MCP server for Dataiku DSS REST APIs, focused on flow analysis and reliable day-to-day operations (projects, datasets, recipes, jobs, scenarios, folders, variables, connections, and code environments).
Cursor one-click install includes placeholder environment values. Update
DATAIKU_URL,DATAIKU_API_KEY, and optionallyDATAIKU_PROJECT_KEYafter adding the server.
What You Get
- Deterministic normalized flow maps (
project.map) with recipe subtypes and connectivity. - Summary-first outputs with explicit raw/detail toggles where needed.
- Broad test coverage (unit + live integration + optional destructive integration suite).
- Strong error taxonomy in responses:
not_found,forbidden,validation,transient,unknownwith retry hints.
Tool Coverage
project:list,get,metadata,flow,mapdataset:list,get,schema,preview,metadata,download,create,update,deleterecipe:list,get,create,update,delete,downloadjob:list,get,log,build,buildAndWait,wait,abortscenario:list,run,status,get,create,update,deletemanaged_folder:list,get,contents,download,upload,delete_filevariable:get,setconnection:infercode_env:list,get
Prerequisites
- Node.js 20+
- npm
- Dataiku DSS URL + API key
Quick Start
Run as a local CLI after build:
Use directly from npm (after publish):
Local Build And Testing
Recommended local workflow from repo root:
Optional live DSS integration tests:
Repository Layout
src/: MCP server and tool implementations.tests/: unit + integration test suites.examples/: demos, fixtures, artifacts, and ad-hoc local scripts.bin/: package executable entrypoint.dist/: compiled output (generated).
Create a local env file:
Run directly in dev:
Example scripts and sample outputs are kept under examples/ to avoid root-level clutter.
Environment Variables
DATAIKU_URL: DSS base URLDATAIKU_API_KEY: DSS API keyDATAIKU_PROJECT_KEY(optional): default project keyDATAIKU_REQUEST_TIMEOUT_MS(optional): per-attempt request timeout in milliseconds (default:30000)DATAIKU_RETRY_MAX_ATTEMPTS(optional): max attempts for retry-enabled requests (GETonly, default:4, cap:10)DATAIKU_DEBUG_LATENCY(optional): set to1/trueto include per-tool timing diagnostics instructuredContent.debug.latency(off by default)
MCP Client Setup Guide
Use this server command in clients (npm package):
Windows note: if your MCP client launches commands without a shell, use npx.cmd:
You can also run TypeScript directly during development:
Claude Desktop
- Open Claude Desktop ->
Settings->Developer->Edit Config. - Add this under
mcpServersinclaude_desktop_config.json:
Cursor
Cursor supports both project-scoped and global MCP config:
- Project:
.cursor/mcp.json - Global:
~/.cursor/mcp.json
Example:
Cline (VS Code extension)
- Open Cline -> MCP Servers -> Configure MCP Servers.
- Add this server block in
cline_mcp_settings.json:
Codex / project-level MCP config
This repo already includes a project-scoped MCP file at .mcp.json.
The checked-in .mcp.json uses node node_modules/tsx/dist/cli.mjs src/index.ts for cross-platform startup (including Windows); run npm ci first.
NPM Release Workflow
This repo includes a manual GitHub Actions release workflow:
- Workflow file:
.github/workflows/release.yml - Trigger:
Actions->Release NPM Package->Run workflow
Inputs:
bump:patch | minor | majorversion: optional exact version (overridesbump)publish: whether to publish to npm
Required repository configuration:
- GitHub variable:
NPM_RELEASE_ENABLED=true - Optional variable:
NPM_PUBLISH_ACCESS=public - Trusted publisher configured on npmjs.com for this package/repo/workflow
The workflow will:
- Install dependencies, run checks/tests, and build.
- Bump package version and create git tag.
- Push commit + tag to
main. - Publish to npm with GitHub OIDC trusted publishing (if
publish=true). - Create a GitHub Release with generated notes.
Trusted publishing setup (npm):
- Open
https://www.npmjs.com/package/dataiku-mcp->Settings->Trusted Publisher. - Choose
GitHub Actions. - Set:
- Organization or user:
clssck - Repository:
Dataiku_MCP - Workflow filename:
release.yml
- Organization or user:
- Save.
Official MCP Registry
This repo is configured for MCP Registry publishing:
- Metadata file:
server.json - Workflow:
.github/workflows/publish-mcp-registry.yml - Required package field:
mcpNameinpackage.json
Server namespace:
io.github.clssck/dataiku-mcp
Publish paths:
- Manual: run
Publish to MCP Registryin GitHub Actions. - Automatic: run the npm release workflow with
publish=true(it triggers MCP Registry publish).
Validation notes:
server.json.namemust matchpackage.json.mcpName.server.json.packages[].identifier+versionmust reference a real npm publish.
Recommended Verification Prompt
After adding the server in a client, run:
projectwith{ "action": "map", "projectKey": "YOUR_PROJECT_KEY" }(defaults tomaxNodes=300,maxEdges=600; override as needed)
You should receive a flow summary in text and normalized nodes, edges, stats, roots, and leaves under structuredContent.map.
When truncation limits are applied (default maxNodes=300, maxEdges=600), structuredContent.truncation reports before/after node+edge counts and whether truncation occurred.
Notes
project.mapreturns a compact text summary; full normalized graph is instructuredContent.map.- Arrays in normalized map output are deterministically sorted to reduce diff churn.
job.waitandjob.buildAndWaitincludestructuredContent.normalizedStatewith one ofterminalSuccess | terminalFailure | timeout | nonTerminalwhile preserving raw DSSstate.- With
DATAIKU_DEBUG_LATENCY=1, responses include per-tool and per-API-call latency metrics understructuredContent.debug.latency. - List-style responses are token-bounded by default; use
limit/offset(and action-specific caps likemaxNodes,maxEdges,maxKeys,maxPackages) to page or expand results when needed. dataset.getandjob.getare summary-first by default; passincludeDefinition=trueto include full DSS JSON instructuredContent.definition.
Sources
- MCP local server connection docs: https://modelcontextprotocol.io/docs/develop/connect-local-servers
- Cursor MCP docs: https://cursor.com/docs/context/mcp
- Cline MCP docs: https://docs.cline.bot/mcp/configuring-mcp-servers
來源:README.md,提交 1d20e26
工具
0版本歷史
1- v1.1.1最新Sep 16, 2026
評分
已滿足適用檢查項的 50%
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25 分
請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。
- 已宣告認證方式10 分
請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。
- 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