How do I run a Xiaohongshu MCP on a cloud server without a browser?

Take the browser out of the picture. Self-hosted Xiaohongshu MCP servers run a full Chrome inside the container, logged into your account, and that is where cloud deployments go wrong: searches that hang until a 60-second deadline, containers that shut down when idle, a login that has to be redone by QR code, and a datacenter IP that Xiaohongshu treats with more suspicion than a home connection. The project's issue tracker has reports of exactly these. With a hosted server, your machine makes one HTTPS request and gets JSON back. That runs fine on a 1 GB VPS, a cron job in GitHub Actions or a serverless function.

Two ways to wire it

• Your agent speaks MCP. Add https://api.okdata.online/mcp as a remote server with an Authorization: Bearer header. The agent gets find, describe and call, and can reach every platform, not only Xiaohongshu. Client configs are on the MCP page.
• A scheduled job, no agent. Call the HTTP endpoints directly: Xiaohongshu note search and note detail. The script below is what you would put in cron.

import os, time, uuid, requests

API = "https://api.okdata.online"
H = {"Authorization": f"Bearer {os.environ['OKDATA_API_KEY']}"}

def search(query, sort="latest", tries=3):
    key = str(uuid.uuid4())  # same key on every retry: never billed twice
    for _ in range(tries):
        r = requests.get(API + "/v1/xiaohongshu/notes/search",
                         headers={**H, "Idempotency-Key": key},
                         params={"query": query, "sort": sort,
                                 "fields": "id,url,title,metrics.likes,published_at"},
                         timeout=90)
        body = r.json()
        if body["code"] == "ok":
            return body["data"]["items"]
        if body["retry"] in ("yes", "same_key"):
            time.sleep(body.get("retry_after_seconds") or 5)
            continue
        raise RuntimeError(f"{body['code']}: {body['message']}")
    raise RuntimeError("gave up")

for note in search("露营装备"):
    print(note["metrics"]["likes"], note["title"], note["url"])

What it costs to run on a schedule

Job	Calls	Cost
20 keywords, first page, once a day	20 a day	$0.40 a day
Same job for 30 days	600	$12.00
Plus full text of 10 new notes a day for 30 days	300	$6.00

There is nothing to pay for the server beyond what you already run, since it only waits on HTTP.

Gotchas

• Set your HTTP timeout to 60 seconds or more. Most calls are quicker, but a slow platform response should not be cut off on your side.
• A timeout from us comes back as timeout with charged: null. Resend with the same Idempotency-Key and you will get the stored result or a fresh run, never a second charge. The errors page lists every code.
• Watch X-RateLimit-Remaining if you run many keywords in parallel; a rate_limited reply says how many seconds to wait.
• Xiaohongshu hides view counts, so metrics.views is always null.

When to keep the self-hosted MCP

If your agent has to publish notes or reply to comments as your account, the browser-based server is still the tool, and a machine on a home connection tends to behave better than a datacenter VM. Keep it for writing and move reading to an API key. For one-off research by hand, the Xiaohongshu app is still the cheapest way to look.