OpenVidu MCP servers#
An MCP server gives a coding agent tools it can call. OpenVidu provides the documentation server below, included in the OpenVidu Agent Plugin. Add it on its own when your client doesn't load plugins, or when you would rather not install one.
OpenVidu documentation server#
- Name:
openvidu-docs - URL:
https://docs-mcp.openvidu.io/mcp - Transport: Streamable HTTP, with no account or API key
It searches and reads the OpenVidu Meet and OpenVidu Platform documentation of every release from 3.4, and points your agent to LiveKit's documentation when it needs more, such as the SDK references. Your agent calls its tools when a task needs them:
| Tool | What it does |
|---|---|
search_docs |
Searches the documentation, linking each result to the section that matched. |
get_doc_page |
Reads one or more pages, whole or a section at a time. |
list_doc_sections |
Lists the documentation's sections and their pages. |
list_versions |
Lists the documentation versions, and the one used by default. |
resolve_openvidu_version_edition_product |
Explains how to find out the version, edition and product a project uses, and maps a LiveKit Server version to OpenVidu's. |
get_changelog |
Returns the release notes of a version. |
get_pricing_info |
Returns the editions, plans and pricing. |
The tools that read the documentation take the version your deployment reports (3.9.1) and answer from that release's documentation (3.9); without one, they use the latest. A version the server doesn't carry is an error, never an answer for another release. There is nothing to update on your side: the server is rebuilt each time this documentation is published.
Add it to your client#
Add --scope project to share it with your team through the repository's .mcp.json, or --scope user to have it in every project.
In Customize → Connectors, select +, then Add custom connector, and paste the server URL:
On Team and Enterprise plans, an owner adds it for the whole organization, from Organization settings → Connectors.
Or add it to ~/.cursor/mcp.json (every project) or .cursor/mcp.json (this project):
Or, for this workspace only, in .vscode/mcp.json:
In the IDE extension: the gear menu, MCP servers → Add server, with the Streamable HTTP type.
Leave out -s user to add it to the current project only.
Or add it to ~/.kiro/settings/mcp.json (every project) or .kiro/settings/mcp.json (this project):
Any client that speaks MCP over Streamable HTTP works with the URL alone. The usual configuration is:
{
"mcpServers": {
"openvidu-docs": {
"type": "http",
"url": "https://docs-mcp.openvidu.io/mcp"
}
}
}
Check your client's documentation for where that file lives, and whether it names the field type, transport or nothing at all.
To check that it answers, ask your agent which OpenVidu documentation versions it can read.
Privacy#
Each request your coding agent makes to docs-mcp.openvidu.io is logged as one line: the tool it called, what it asked for (for a search, the search terms), the documentation version, whether it worked, how long it took, and the agent's name and version when it sends them. What was asked is kept to learn what developers look for and what the documentation is missing. Your conversation, your prompts and your code are never sent to us or stored.
Your IP address is used only to group one client's requests into a visit, and is discarded before anything is written to storage: visits are labelled with a random identifier that cannot be traced back to an address or linked to a later visit. Request logs are deleted after 7 days. The archive, which contains no addresses, is deleted after 395 days so a year-over-year comparison is possible. Aggregate counts are kept. Nothing is shared with third parties.
When your agent reads LiveKit's documentation, it fetches those pages from docs.livekit.io itself: that request never passes through OpenVidu, and LiveKit's own policies apply to it. The rest of what openvidu.io collects is in the privacy policy.
Troubleshooting#
The client shows the server as failed or offline
The transport must be HTTP (Streamable HTTP), not sse or stdio, and the URL must end in /mcp. To check the server from a shell, list its tools:
Opening the server URL in a browser gives an error
That is expected: the server only accepts POST requests. It is not a web page, and for the same reason a browser-based MCP client cannot use it.
A page you know exists isn't found
The server is rebuilt when this documentation is published, so a page published minutes ago may not be there yet.
Without MCP#
Every page of this site is also published as Markdown: add index.md to a page's URL, as in https://openvidu.io/latest/docs/getting-started/index.md. https://openvidu.io/llms.txt indexes them for the latest release, and https://openvidu.io/<version>/llms.txt for each version, following the llms.txt convention. Where your agent supports MCP, the server is the better option: it searches, and it answers for the version you ask for.