srgplus-mcp
MCP server for SRG+ — lets Claude (and any MCP-aware agent) manage hubs, channels, content, assets, users, and workspaces through the SRG+ API.
README
srgplus-mcp
<!-- MCP Registry ownership proof — required by mcp-publisher for the io.github.srgplus/* namespace. Do not remove. --> mcp-name: io.github.srgplus/srgplus-mcp
MCP server for SRG+ — lets Claude (and any MCP-aware agent) manage hubs, channels, content, assets, users, and workspaces through the SRG+ API.
Two ways to run it:
- Hosted HTTP (recommended for production / claude.ai web / Cursor / Cline) — single endpoint, header-based auth, multi-tenant
- Local stdio (for desktop dev / offline) — single user, env-var auth, runs as a child process of the agent
Both modes share the same tools and the same SDK underneath — you pick the transport that fits your client.
Hosted HTTP
Run the server
pip install 'srgplus-mcp[server]'
srgplus-mcp-serve # listens on $PORT (default 8090)
Or in Docker:
docker build -t srgplus-mcp .
docker run -p 8090:8090 srgplus-mcp
Health check:
curl http://localhost:8090/health
Connect from Claude / Cursor / any MCP client
{
"mcpServers": {
"srgplus": {
"url": "http://localhost:8090/mcp",
"headers": { "X-API-Key": "srgplus_your_key_here" }
}
}
}
Or use Authorization: Bearer srgplus_... instead of X-API-Key — both work.
For a hosted public endpoint pointed at your SRG+ workspace, the URL becomes
https://mcp.srgplus.com/mcp (rolling out — see SRGDEV-8 follow-ups for the
deploy plan).
OAuth (claude.ai web, Connectors Gallery, ChatGPT Apps)
For browser-based clients that require OAuth 2.1 (DCR + PKCE):
- In claude.ai → Settings → Connectors → Add custom connector
- URL:
https://mcp.srgplus.com/mcp - Leave OAuth fields empty — the wizard will discover, DCR-register, and redirect to a consent page
- On the consent page: paste your SRG+ workspace API key
- Tools appear in claude.ai
Endpoints exposed (all under the same hostname as /mcp):
| Endpoint | Purpose |
|---|---|
/.well-known/oauth-authorization-server |
RFC 8414 metadata |
/.well-known/oauth-protected-resource |
RFC 9728 metadata |
/oauth/register |
RFC 7591 Dynamic Client Registration |
/oauth/authorize |
Authorization endpoint with consent page |
/oauth/token |
Token endpoint (PKCE S256 required) |
/oauth/revoke |
RFC 7009 revocation |
Backwards compatibility: the X-API-Key header and Authorization: Bearer srgplus_... flows from above still work. OAuth tokens (RFC-shaped
JWTs) are detected automatically when the bearer value doesn't start with
srgplus_.
Multiple workspaces
Each SRG+ workspace API key is scoped to a single workspace, so one OAuth session today connects one workspace.
To connect Claude to several workspaces at once, add the SRG+ connector multiple times in claude.ai → Settings → Connectors. Each instance goes through its own OAuth flow with the API key for that workspace:
SRG+ (Acme) → https://mcp.srgplus.com/mcp → API key for Acme workspace
SRG+ (Personal) → https://mcp.srgplus.com/mcp → API key for Personal workspace
SRG+ (Studios) → https://mcp.srgplus.com/mcp → API key for Studios workspace
Tools from each connector show up in claude.ai under their connector name, so you can scope a request to a specific workspace by mentioning the connector ("use SRG+ Acme to find content X").
A single-OAuth multi-workspace flow (Airtable-style: pick multiple workspaces in one consent screen, server provisions scoped keys automatically) is tracked in SRGDEV-27 and will land in a future release. Until then, the multi-connector approach above covers the same use case with no extra code.
Production secrets
In production, set these via GCP Secret Manager so token state survives process restarts:
| Env var | Purpose |
|---|---|
OAUTH_ISSUER |
Canonical issuer URL (e.g. https://mcp.srgplus.com) |
OAUTH_CLIENT_REGISTRATION_KEY |
HMAC key for DCR client_id JWTs |
OAUTH_TOKEN_SIGNING_KEY |
HMAC key for codes/access/refresh tokens |
OAUTH_API_KEY_ENCRYPTION_KEY |
32-byte AES-GCM key (base64url) for wrapping the SRG+ workspace api_key inside JWTs |
If any are unset the server generates an ephemeral random key per process — fine for local development, but every process restart invalidates all in-flight authorizations.
How auth works
Each request must carry the workspace API key in either header:
X-API-Key: srgplus_...Authorization: Bearer srgplus_...
The server doesn't pre-validate the key — it binds it to the SDK's
per-request contextvar via SRGClient.use_api_key(...) and lets the SRG+
SDK make the actual call. Bad keys surface as 401 from the upstream API on
the first tool invocation.
A single shared SRGClient (and therefore a single httpx connection pool)
serves every request, so the per-request overhead is just a contextvar
set/reset — no per-key client construction or cache.
Local stdio (developer mode)
Use this when you're building locally against SRG+ and want a child-process MCP without running an HTTP server.
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"srgplus": {
"command": "uvx",
"args": ["srgplus-mcp"],
"env": {
"SRG_API_KEY": "srgplus_your_key_here"
}
}
}
}
Restart Claude Desktop. The uvx command downloads and runs the package
automatically — no separate install step needed.
Claude Code
claude mcp add srgplus -- uvx srgplus-mcp
export SRG_API_KEY=srgplus_your_key_here
Getting an API key
Log in to SRG+ → Settings → Workspaces → Select Workspace → API Keys →
create a new key. Use the same key for both stdio (SRG_API_KEY env var) and
hosted HTTP (X-API-Key header).
Available tools
Hub Profiles
list_hub_profiles · list_managed_hub_profiles · get_hub_profile · get_hub_profile_by_username · filter_hub_profiles · create_hub_profile · update_hub_profile · archive_hub_profile · restore_hub_profile · delete_hub_profile · join_hub_profile · invite_to_hub_profile · list_invitations · update_invitation · delete_invitation · get_invitation_link · move_hub_profile_to_workspace · turn_on_hub_profile_community
Channels
list_channels · get_channel · get_channel_by_name · create_channel · update_channel · archive_channel · delete_channel · create_category · update_category · archive_category · delete_category · get_category_by_slugs · create_section · update_section · delete_section
Contents
list_contents · get_content · get_content_v2 · create_content · update_content · move_content · search_contents · add_content_to_category · add_content_to_categories · remove_content_from_categories · get_category_references · create_content_section · update_content_section · delete_content_section · add_subcontent · get_subcontent · move_subcontent · delete_subcontent · patch_content_progression · get_progression_stats
Assets
upload_asset · list_assets · get_asset · search_assets · create_embed_asset · update_asset · patch_media_progression
upload_assetcreates an asset and uploads its bytes in one call (from asource_urlorbase64_content) — use it for any file, image, or video. The oldercreate_image_asset/create_video_asset/create_file_asset/create_media_assettools only register an empty record and are deprecated.
Users & Permissions
get_user · check_user_exists_by_email · check_user_exists_by_phone · get_workspace_users · give_permission · delete_permission · can_read · can_edit · can_archive · can_create_child · can_manage_permissions · is_member · get_permission_targets · list_permission_groups · get_permission_group · create_permission_group · update_permission_group · delete_permission_group · add_users_to_permission_group · remove_user_from_permission_group
Workspace
get_workspace · update_workspace · get_workspace_hub_profiles · list_workspace_actions · get_workspace_action · create_workspace_action · update_workspace_action · delete_workspace_action · invite_to_workspace · get_workspace_invitation_link · list_workspace_invitations · update_workspace_invitation · delete_workspace_invitation
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。