srgplus-mcp
Official# 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+](https://srgplus.com) — 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
```bash
pip install 'srgplus-mcp[server]'
srgplus-mcp-serve # listens on $PORT (default 8090)
```
Or in Docker:
```bash
docker build -t srgplus-mcp .
docker run -p 8090:8090 srgplus-mcp
```
Health check:
```bash
curl http://localhost:8090/health
```
### Connect from Claude / Cursor / any MCP client
```json
{
"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):
1. In claude.ai → **Settings → Connectors → Add custom connector**
2. URL: `https://mcp.srgplus.com/mcp`
3. Leave OAuth fields empty — the wizard will discover, DCR-register, and
redirect to a consent page
4. On the consent page: paste your SRG+ workspace API key
5. 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](https://sergecreator.atlassian.net/browse/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`:
```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
```bash
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_asset` creates an asset and uploads its bytes in one call (from a `source_url` or `base64_content`) — use it for any file, image, or video. The older `create_image_asset` / `create_video_asset` / `create_file_asset` / `create_media_asset` tools 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`
TDQS
Scored across 103 tools
Most tools have distinct purposes, but the high number of similar CRUD operations (e.g., create_hub_profile vs create_channel vs create_category, get_channel_by_name vs get_channel) could cause confusion. The descriptions are detailed enough to differentiate, though some overlap exists between content/category/section operations.
The naming convention is mostly consistent with verb_noun patterns (list_, get_, create_, update_, delete_, etc.), but there are some deviations like `can_read`, `can_edit`, `is_member`, and `get_srgplus_guide` that don't fit the typical CRUD pattern. The mix of `list_*` and `get_*` for similar operations is slightly inconsistent.
With 103 tools, this is a very large surface area. While the domain is broad (workspaces, hub profiles, channels, categories, sections, content, assets, users, permissions, invitations, automation actions), the sheer number makes navigation difficult. Many tools are near-duplicates (e.g., multiple invitation link generators, multiple permission checkers), suggesting the count could be trimmed.
The tool set appears to cover the full lifecycle of the platform: workspace management, hub profile management, content hierarchy (channel→category→section→content), asset upload, user/permission management, invitations, automation actions, and permission groups. Missing features like content deletion are intentional (archiving instead), and the guide tool fills knowledge gaps.