seerr-mcp

A remote MCP server that lets an AI assistant search your media catalog and add movies and shows to Overseerr or Jellyseerr. Ask for a title in plain language, the assistant finds it, confirms the match, and requests it. Radarr and Sonarr take it from there.

It runs as a single Cloudflare Worker with no database and no build step. Any MCP client that speaks Streamable HTTP can connect: Claude Code, Claude Desktop, a hosted voice agent that answers a phone number, or your own code.

MCP client  ->  POST /mcp on your Worker      (bearer gated by MCP_TOKEN)
            ->  Overseerr / Jellyseerr        (/api/v1, X-Api-Key)
            ->  Radarr / Sonarr add

Tools

Related MCP server: Overseerr MCP Server

What you need

• A running Overseerr or Jellyseerr that this Worker can reach over the public internet, plus an admin api key from Settings, General, API Key.

• A Cloudflare account. The free plan is enough.

• Node 18 or newer, for npx wrangler.

The api key should belong to an admin. Requests then approve themselves, and the availability lookups that stop duplicate requests are allowed to run.

Setup

git clone https://github.com/Neoo-Blue/seerr-mcp.git
cd seerr-mcp
npm install

Generate a bearer token. This is the only thing standing between the open internet and write access to your media stack, so make it a real random value and keep it out of shell history you share:

openssl rand -hex 24

Set the three secrets. Wrangler prompts for each value and never writes it to disk:

npx wrangler secret put MCP_TOKEN      # the value you just generated
npx wrangler secret put SEERR_URL      # e.g. https://seerr.example.com
npx wrangler secret put SEERR_APIKEY   # Seerr Settings, General, API Key

Deploy:

npx wrangler deploy

You get back a url like https://seerr-mcp.your-subdomain.workers.dev. That is already public https, so it works as is. To serve it from your own hostname instead, uncomment the routes block in wrangler.jsonc and set your zone.

Check it works

export SEERR_MCP=https://seerr-mcp.your-subdomain.workers.dev
export MCP_TOKEN=the-token-you-generated

# health, no auth needed
curl $SEERR_MCP/

# list the tools
curl -s -X POST $SEERR_MCP/mcp \
  -H "Authorization: Bearer $MCP_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

# search the catalog
curl -s -X POST $SEERR_MCP/mcp \
  -H "Authorization: Bearer $MCP_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"search_media","arguments":{"query":"Dune"}}}'

A wrong token returns unauthorized with a 401. If tools/list works but search_media says it could not reach the library, the Worker cannot see your SEERR_URL or the api key is wrong.

Connect a client

Claude Code

claude mcp add --transport http seerr https://seerr-mcp.your-subdomain.workers.dev/mcp \
  --header "Authorization: Bearer YOUR_MCP_TOKEN"

Then just ask: "add Dune to the library" or "is The Bear available yet".

Claude Desktop and other MCP clients

Add a remote MCP server with the url https://your-worker/mcp and an Authorization: Bearer YOUR_MCP_TOKEN header. Clients that only speak stdio can reach it through mcp-remote:

{
  "mcpServers": {
    "seerr": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://seerr-mcp.your-subdomain.workers.dev/mcp",
        "--header", "Authorization: Bearer YOUR_MCP_TOKEN"
      ]
    }
  }
}

A voice agent on a phone number

The same three tools turn this into a phone line people can call to add a movie. See docs/voice-agent.md for the wiring and a ready made concierge prompt.

Local development

cp .dev.vars.example .dev.vars   # fill in the three values
npm run dev

.dev.vars is gitignored. Point your curl calls at http://localhost:8787.

Configuration

Notes on behaviour

• Overseerr accepts repeat requests for titles that are already available or already downloading, with no error, which quietly creates duplicate rows. request_media therefore checks real availability first and only posts a request when the title is genuinely absent. If it is already there, it says so instead of pretending it started a download.

• search_media returns the tmdb_id in its text so the model can pass it to request_media, and the tool description tells the model never to read that number out loud. Worth keeping if you edit the prompts.

• The Worker never exposes your Seerr UI. It only calls /api/v1 server side, so SEERR_URL and the api key stay on Cloudflare and never reach the client.

Security

Anyone holding MCP_TOKEN can add anything to your library. Treat it like a password: rotate it with npx wrangler secret put MCP_TOKEN if it ever lands in a chat log, a screenshot, or a repo. Set SEERR_AUTOAPPROVE to 0 if you would rather approve each request by hand.

License

MIT. See LICENSE.