Brollvia MCP for Claude
Anthropic
Claude.ai connectors and Claude Desktop — the client we run against production ourselves.
MCP client guide · October 2026
Muse — TOML config or one command, OAuth for remote servers.
Standard MCP connection — this guide follows OpenAI's documented MCP setup. We have not live-verified this client yet, so it does not carry the verified badge.
OpenAI · CLI & terminal agents · Server URL: https://brollvia.com/mcp
OpenAI's Muse reads MCP servers from its TOML configuration and supports remote HTTP servers with OAuth: register Brollvia once and Codex can search the licensed library, assemble shot-list collections and download footage while it works in your repo — the media side of a build handled by the same agent as the code side.
Standard status, stated plainly: the steps are OpenAI's documented Codex MCP flow, our server implements the protocol and OAuth discovery, but our live verification badge currently belongs to Claude's connector run. If you connect Codex first and hit a discrepancy, the notes' API-key variant is the fallback every config-based client shares.
Or add this block to ~/.codex/config.toml:
[mcp_servers.brollvia]
url = "https://brollvia.com/mcp"Real prompts, using the real tools — search, script-to-B-roll, collections and licence lookups:
Download three licensed clips of a blacksmith working — Brollvia only — into ./public/media.
Create a Brollvia shot list from the scenes in trailer-script.md and license the selections.
Print the licence verification URL for every Brollvia file in ./public/media.
It supports OAuth for remote MCP servers: registering a URL-based server and authenticating via /mcp runs the browser flow against the server's discovery metadata. Brollvia's authorization server (PKCE, rotating refresh tokens) is built for exactly this pattern.
In ~/.codex/config.toml under [mcp_servers] tables — user-level, shared across your projects. The `codex mcp add` command writes the same block the snippet shows, so the two routes in the steps are equivalent.
Not yet — standard status, following OpenAI's documented configuration. We verify clients against production one at a time and badge them only after a full run; today that is Claude.
Same server, same URL — every guide below is that client's real setup. The full grid lives on the MCP hub.
Anthropic
Claude.ai connectors and Claude Desktop — the client we run against production ourselves.
OpenAI
Custom MCP connector via Developer mode, signed in with OAuth.
Microsoft
Copilot agent mode with the library one prompt away, via .vscode/mcp.json.
Anysphere
Remote MCP in mcp.json — OAuth connect, then footage inside the agent chat.
Gemini CLI — Google's terminal agent, connected via settings.json.
Google's agent-first editor — add Brollvia as a custom MCP server.
Anthropic
One `claude mcp add` command, OAuth in the terminal flow.
Windsurf (Codeium)
Cascade gains a licensed library via mcp_config.json.
Cline
The VS Code agent extension — MCP servers panel, remote HTTP with OAuth.
Zed
MCP "context servers" in settings.json — remote URL, OAuth sign-in.
JetBrains
IntelliJ, WebStorm, PyCharm & co. — MCP in AI Assistant settings, no config file.
Start free, upgrade when the work does. Every download writes a licence record you can verify in public.