Productboard MCP — Curve & Flex
Enables reading and writing Productboard data via MCP tools, with Google OAuth login, per-user allowlist, and workspace binding for secure, audited access.
README
Productboard MCP — Curve & Flex
Custom MCP server exposing read/write tools over the Productboard v2 API, with Google-OAuth 2.1 login, a per-email allowlist, and per-user workspace binding (Curve + Flex served from one Cloud Run service).
This is the canonical source. Baseline captured from the source of the
May 26 2026 production deploy and committed as v1 (July 2026). Earlier local
prototypes live in ../_Archive/ — do not edit those; they are pre-OAuth and
out of date.
Layout
src/server.py— the MCP tools (the tool surface Claude sees)src/productboard.py— Productboard v2 API client (auth, retries, pagination, notes)src/auth.py— the MCP's own OAuth 2.1 (Google IDP + JWT). NOT Productboard auth.src/allowlist.py— who may use the MCP + their workspace bindingDockerfile,requirements.txt,deploy.sh— build & deployROADMAP.md— forward plan
Deploy
./deploy.sh # READ_ONLY=true (write tools disabled) — default
READ_ONLY=false ./deploy.sh # enable write tools
Target: project productboard-mcp, region us-central1, service
productboard-mcp (gcloud run deploy --source=.). Secrets — the per-workspace
Productboard API tokens, the OAuth client secret, and the JWT signing key — are
mounted from Secret Manager. Never store them here.
Rules
- No secrets in the repo. Tokens live in Secret Manager;
src/allowlist.pyis the only access config in-tree, and git history is its audit trail. - Safe surface by design: no create-feature tool, no permanent delete (soft-delete
via
archived). Keep it that way. - Write access is per-user: only allowlist entries with
writer: Truecan call the write tools, even when the service runsREAD_ONLY=false(the env var stays as the global kill switch). Currently: Luke only.
Revocation
- One user, immediately: set
tokens_valid_afteron their allowlist entry (or remove the entry), commit, redeploy. Takes effect on their next request — permissions are re-derived from the allowlist on every call, not frozen into tokens. - Everyone at once (break-glass): bump the
TOKEN_VERSIONenv var (gcloud run services update productboard-mcp --region=us-central1 --update-env-vars=TOKEN_VERSION=<n+1>). All outstanding tokens die instantly; users re-auth via Google. No code change, no key rotation. - Refresh tokens live 7 days; offboarding must still remove the allowlist entry (leaver-checklist item — disabling the Google account alone does NOT cut existing sessions).
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。