Wechat Official Account

We drive the WeChat MP server-side API with curl + jq. Unlike OAuth-bearer connectors, WeChat MP uses a two-step flow:

1. Exchange AppID + AppSecret for an access_token (TTL 7200s, global limit ≈ 2000 calls/day per app).
2. Pass that access_token as a query string parameter on every other call.

The user's credentials are in $WECHAT_APP_ID and $WECHAT_APP_SECRET. Never log or echo $WECHAT_APP_SECRET — treat it like a password.

The WeChat MP API returns standard JSON. Errors are returned with HTTP 200; the body looks like {"errcode": 40013, "errmsg": "invalid appid"}. errcode == 0 means success — show the original errmsg to the user verbatim on any non-zero code.

Important constraints — surface these to the user before they're surprised

• IP whitelist: every API call's source IP must be in this app's IP whitelist (公众平台 → 设置与开发 → 基本配置 → IP 白名单). If you see errcode 40164 ("invalid ip"), the worker's egress IP isn't whitelisted; tell the user to add the IP shown in errmsg and retry.
• Verified account required for publishing: as of 2025-07, only verified (已认证) corporate-subject accounts can call freepublish/* and mass/*. Personal-subject accounts and unverified corporate accounts get a permission error. Drafts (draft/*) and customer messages (message/custom/*) usually work without verification.
• Group-send quota is harsh: 服务号 = 4 sends/month, 订阅号 = 1 send/day. Treat freepublish/submit and mass/sendall like a destructive operation — always confirm with the user before calling them, even if instructions say "publish it". Default to creating a draft and pasting the draft URL.
• Customer-service window is 48 hours: message/custom/send only works for a follower whose openid interacted with the account in the last 48 hours. Outside that window you get errcode 45015.

Recipes

Step 0 — get an access_token (do this first, cache the result)

# Cache to /tmp so subsequent calls in the same session reuse it.
TOKEN_CACHE="/tmp/wx-mp-token-${WECHAT_APP_ID}.json"

# Reuse cached token if it's still valid (we conservatively refresh
# 5 minutes early to avoid edge-of-window failures).
NOW=$(date +%s)
if [ -f "$TOKEN_CACHE" ] && [ "$(jq -r '.exp_at // 0' "$TOKEN_CACHE")" -gt "$((NOW + 300))" ]; then
  WECHAT_ACCESS_TOKEN=$(jq -r '.access_token' "$TOKEN_CACHE")
else
  RESP=$(curl -sS "https://api.weixin.qq.com/cgi-bin/token?grant_type=client_credential&appid=${WECHAT_APP_ID}&secret=${WECHAT_APP_SECRET}")
  WECHAT_ACCESS_TOKEN=$(echo "$RESP" | jq -r '.access_token // empty')
  if [ -z "$WECHAT_ACCESS_TOKEN" ]; then
    echo "Failed to fetch access_token: $RESP" >&2
    exit 1
  fi
  EXPIRES=$(echo "$RESP" | jq -r '.expires_in // 7200')
  jq -nc --arg t "$WECHAT_ACCESS_TOKEN" --argjson e "$((NOW + EXPIRES))" \
    '{access_token:$t, exp_at:$e}' > "$TOKEN_CACHE"
fi
echo "OK token=${WECHAT_ACCESS_TOKEN:0:8}…"

If the response is {"errcode": 40164, ...} (invalid IP) or {"errcode": 40013, ...} (invalid appid) — surface that error to the user; it means the connector setup itself is wrong, not the request.

Verify the connection (always run this once before complex operations)

# Pull basic public-account self-info; cheapest call that proves the token works.
curl -sS "https://api.weixin.qq.com/cgi-bin/account/getaccountbasicinfo?access_token=${WECHAT_ACCESS_TOKEN}" | jq

Upload an image to use inside an article body (returns a public URL)

This is for <img src="..."> tags inside the article HTML. The returned url is hosted by Tencent and works in published articles.

curl -sS -X POST \
  "https://api.weixin.qq.com/cgi-bin/media/uploadimg?access_token=${WECHAT_ACCESS_TOKEN}" \
  -F "media=@/path/to/your-image.jpg"
# → {"url": "http://mmbiz.qpic.cn/..."}

Limits: ≤ 1 MB, JPG/PNG only, no count limit on this endpoint.

Upload a thumbnail (needed as the article cover)

curl -sS -X POST \
  "https://api.weixin.qq.com/cgi-bin/material/add_material?access_token=${WECHAT_ACCESS_TOKEN}&type=thumb" \
  -F "media=@/path/to/cover.jpg" | jq '{media_id, url}'
# → {"media_id": "MEDIA_ID", "url": "http://mmbiz.qpic.cn/..."}

The media_id you get back is what you pass as thumb_media_id when creating a draft. Recommended cover dimensions: 900×500 px, JPG, < 64 KB ideally.

Create a draft (the safe default — do not directly publish)

TITLE="Q1 product update"
AUTHOR="Acme Inc."
THUMB_MEDIA_ID="MEDIA_ID_FROM_PREVIOUS_STEP"
CONTENT_HTML='<p>欢迎关注我们的最新动态。</p><p><img src="http://mmbiz.qpic.cn/..."></p><p>更多内容请见底部「阅读原文」。</p>'
DIGEST="Q1 has been a wild ride — here's what shipped."
SOURCE_URL="https://example.com/q1-recap"   # optional; populates 「阅读原文」, leave empty to omit

PAYLOAD=$(jq -nc \
  --arg title       "$TITLE" \
  --arg author      "$AUTHOR" \
  --arg thumb       "$THUMB_MEDIA_ID" \
  --arg content     "$CONTENT_HTML" \
  --arg digest      "$DIGEST" \
  --arg source_url  "$SOURCE_URL" \
  '{articles: [{
    article_type: "news",
    title: $title,
    author: $author,
    thumb_media_id: $thumb,
    content: $content,
    digest: $digest,
    content_source_url: $source_url,
    need_open_comment: 0,
    only_fans_can_comment: 0
  }]}')

curl -sS -X POST \
  "https://api.weixin.qq.com/cgi-bin/draft/add?access_token=${WECHAT_ACCESS_TOKEN}" \
  -H "Content-Type: application/json; charset=utf-8" \
  --data-raw "$PAYLOAD" | jq
# → {"media_id": "DRAFT_MEDIA_ID"}

Tell the user: "Draft created. Open https://mp.weixin.qq.com → 草稿箱 to review and tap «发表» from there." — that's the safest path because the user controls the publish click in WeChat's own UI.

List drafts (newest first)

curl -sS -X POST \
  "https://api.weixin.qq.com/cgi-bin/draft/batchget?access_token=${WECHAT_ACCESS_TOKEN}" \
  -H "Content-Type: application/json; charset=utf-8" \
  --data-raw '{"offset": 0, "count": 20, "no_content": 1}' \
  | jq '.item[] | {media_id, update_time, title: .content.news_item[0].title}'

Get the full content of one draft

DRAFT_MEDIA_ID="..."
curl -sS -X POST \
  "https://api.weixin.qq.com/cgi-bin/draft/get?access_token=${WECHAT_ACCESS_TOKEN}" \
  -H "Content-Type: application/json; charset=utf-8" \
  --data-raw "$(jq -nc --arg m "$DRAFT_MEDIA_ID" '{media_id: $m}')" | jq

Update an existing draft

curl -sS -X POST \
  "https://api.weixin.qq.com/cgi-bin/draft/update?access_token=${WECHAT_ACCESS_TOKEN}" \
  -H "Content-Type: application/json; charset=utf-8" \
  --data-raw "$(jq -nc --arg m "$DRAFT_MEDIA_ID" --arg t "Updated title" --arg c "<p>new body</p>" --arg th "$THUMB_MEDIA_ID" '
    {media_id: $m, index: 0, articles: {
      title: $t, content: $c, thumb_media_id: $th
    }}')" | jq

Publish a draft to «发表记录» — DESTRUCTIVE, confirm before calling

# Eats one of the monthly publish slots. Always echo back to the user
# what's about to go live and require an explicit "yes" confirmation
# in conversation BEFORE invoking this.
curl -sS -X POST \
  "https://api.weixin.qq.com/cgi-bin/freepublish/submit?access_token=${WECHAT_ACCESS_TOKEN}" \
  -H "Content-Type: application/json; charset=utf-8" \
  --data-raw "$(jq -nc --arg m "$DRAFT_MEDIA_ID" '{media_id: $m}')" | jq
# → {"errcode": 0, "msg_data_id": ..., "publish_id": "..."}

Then poll publish status (publishing is async, takes ~5-30 seconds):

PUBLISH_ID="..."
curl -sS -X POST \
  "https://api.weixin.qq.com/cgi-bin/freepublish/get?access_token=${WECHAT_ACCESS_TOKEN}" \
  -H "Content-Type: application/json; charset=utf-8" \
  --data-raw "$(jq -nc --arg p "$PUBLISH_ID" '{publish_id: $p}')" | jq '{publish_status, fail_idx, article_id, article_url: .article_detail.item[0].article_url}'
# publish_status: 0 = success, 1 = publishing, 2 = original-check-failed,
#                 3 = failed, 4 = published-but-removed, 5 = unverified-removed

When publish_status == 0, return article_detail.item[0].article_url to the user — that's the canonical https://mp.weixin.qq.com/s/... URL.

List already-published articles

curl -sS -X POST \
  "https://api.weixin.qq.com/cgi-bin/freepublish/batchget?access_token=${WECHAT_ACCESS_TOKEN}" \
  -H "Content-Type: application/json; charset=utf-8" \
  --data-raw '{"offset": 0, "count": 20, "no_content": 1}' \
  | jq '.item[] | {article_id, update_time, title: .content.news_item[0].title, url: .content.news_item[0].url}'

Send a customer-service message (one specific follower, 48h window)

OPENID="oXXXXXXXXXXXXXXXXX"   # the recipient follower's openid
TEXT="Hi! Your subscription has been renewed. Thanks for sticking with us."

curl -sS -X POST \
  "https://api.weixin.qq.com/cgi-bin/message/custom/send?access_token=${WECHAT_ACCESS_TOKEN}" \
  -H "Content-Type: application/json; charset=utf-8" \
  --data-raw "$(jq -nc --arg u "$OPENID" --arg t "$TEXT" '
    {touser: $u, msgtype: "text", text: {content: $t}}')" | jq

errcode 45015 = recipient hasn't messaged the account in the last 48h — explain that to the user; there's nothing the API can do about it.

To send an article card via customer message, change to msgtype: "mpnews" with mpnews: {media_id: "..."} (use material/add_news to first create a permanent news media_id).

Pull follower-growth + read stats

BEGIN="2026-04-25"
END="2026-05-01"   # max 7-day window for getusersummary

# New / unsubscribed / cumulative followers per day
curl -sS -X POST \
  "https://api.weixin.qq.com/datacube/getusersummary?access_token=${WECHAT_ACCESS_TOKEN}" \
  -H "Content-Type: application/json; charset=utf-8" \
  --data-raw "$(jq -nc --arg b "$BEGIN" --arg e "$END" '{begin_date: $b, end_date: $e}')" \
  | jq '.list[] | {date: .ref_date, new: .new_user, lost: .cancel_user, source: .user_source}'

# Article reads per day (max 3-day window for getuserread)
curl -sS -X POST \
  "https://api.weixin.qq.com/datacube/getuserread?access_token=${WECHAT_ACCESS_TOKEN}" \
  -H "Content-Type: application/json; charset=utf-8" \
  --data-raw "$(jq -nc --arg b "$BEGIN" --arg e "$BEGIN" '{begin_date: $b, end_date: $e}')" | jq

datacube/* requires an authenticated (已认证) account with the data-cube permission.

Manage the bottom menu

# Read current menu
curl -sS "https://api.weixin.qq.com/cgi-bin/menu/get?access_token=${WECHAT_ACCESS_TOKEN}" | jq

# Replace the menu (max 3 top-level buttons; each top-level button can have ≤ 5 sub-buttons)
MENU=$(jq -nc '{
  button: [
    {type: "click", name: "今日推荐", key: "DAILY_REC"},
    {name: "更多",
     sub_button: [
       {type: "view", name: "官网",   url: "https://example.com"},
       {type: "view", name: "联系我们", url: "https://example.com/contact"}
     ]}
  ]
}')
curl -sS -X POST \
  "https://api.weixin.qq.com/cgi-bin/menu/create?access_token=${WECHAT_ACCESS_TOKEN}" \
  -H "Content-Type: application/json; charset=utf-8" \
  --data-raw "$MENU" | jq
# → {"errcode": 0, "errmsg": "ok"}

Get the follower list (paginated by next_openid)

NEXT=""
curl -sS "https://api.weixin.qq.com/cgi-bin/user/get?access_token=${WECHAT_ACCESS_TOKEN}&next_openid=${NEXT}" \
  | jq '{total, count, openids: .data.openid, next: .next_openid}'
# Loop: pass the returned `next_openid` as NEXT until count < 10000.

For each openid, batch-fetch profile (≤ 100 per call):

curl -sS -X POST \
  "https://api.weixin.qq.com/cgi-bin/user/info/batchget?access_token=${WECHAT_ACCESS_TOKEN}" \
  -H "Content-Type: application/json; charset=utf-8" \
  --data-raw '{"user_list": [{"openid": "OPENID1", "lang": "zh_CN"}]}' \
  | jq '.user_info_list[] | {openid, nickname, subscribe_time, tagid_list}'

HTML rules for article content

The content field in draft/add accepts a subset of HTML, not full web HTML:

• Allowed: <p>, <span>, <strong>, <em>, <a href>, <img src>, <br>, <h1>–<h3>, <ul>/<ol>/<li>, <blockquote>, <section>.
• Inline styles only (<p style="...">) — no <style> blocks, no <link> to external CSS.
• All <img> src URLs must be either previously-uploaded mmbiz.qpic.cn URLs (from media/uploadimg) or already-published mmbiz URLs. WeChat strips images hosted on third-party domains.
• Total content size limit: ~ 20 K characters (HTML included).

Common errcode cheat-sheet

Why we use bare curl + jq (not an SDK)

Skill bodies must be self-contained shell. Calling pip install wechatpy is not allowed in the sandbox, and the API surface here is small (15-ish endpoints) and stable since 2018. The python SDKs (wechatpy, werobot) are useful references for parameter shapes, but every recipe above is a direct call to the documented endpoint — the SDKs are just thin wrappers around the same JSON.