以下が実際に Claude Code とCodex で明示的に指示して、openai-responses-mcp の動作を確認した結果です。
Codex での呼び出し例 :
user
openai mcp を使って今日の最新の東京の天気を教えてください。とても急いでいるので精度よりスピードを重視してください。クイックに回答してほしい
codex
MCPで「東京の今日の天気」を高速検索します。結果をすぐ要約して返します。
tool running...
openai-responses.answer_quick({"query":"Tokyo weather today"})
tool success, duration: 23.14s
openai-responses.answer_quick({"query":"Tokyo weather today"}){"answer": "Tokyo weather today (2025-08-19 JST):\n- As of 13:57, it's mostly sunny around 34°C (93°F). \n- Forecast for the rest of today: partly sunny,hot and humid. High 34°C (93°F), low 26°C (80°F). Stay hydrated if outdoors. \n\n- Key points:\n - Date/time (Tokyo): 2025-08-19 13:57 JST. \n - Currentconditions: Mostly sunny, ~34°C (93°F). \n - Today's range: High 34°C/93°F, Low 26...thinkingInterpreting weather dataI received a structured JSON that indicates the date is 2025-08-19 and that the weather is mostly sunny at 34°C—perfect, since it matches today. However,there are no citations provided. The user's request prioritizes speed, so I can deliver this info in Japanese quickly. While I didn't perform a recencycheck as per AGENTS.md guidelines, I can note the time as 13:57 JST. I'll structure my response in bullet points and keep it concise. No fabrications here,just the tool's output!Delivering quick weather updateI see that the user just wants a quick weather update, so I can ask a follow-up question about hourly breakdowns: "時間ごとの内訳が必要ですか?", but Ishould keep it simple. The environment restricts some features, but the tool worked, so that's good. I'll avoid heavy formatting as per CLI guidelines anduse bullet points instead.Here's a quick summary in Japanese:- 今日の東京の天気(2025-08-19 JST)- 現在: 晴れがち、約34°C- 予報: 午後は所により晴れ、蒸し暑い。最高34°C、最低26°C- 注意: こまめな水分補給- 出典: 更新時刻: 13:57 JSTcodex- 今日の東京(2025-08-19 13:57 JST 時点)- 現在: 晴れがち、約34°C- 本日予報: 部分的に晴れ、暑く湿度高め- 気温: 最高 34°C / 最低 26°C- メモ: 炎天下の外出はこまめな水分補給を時間別や雨アラートが必要ならお知らせください。
Flux からこのAPIを呼び出す際には Content-Type を application/x-www-form-urlencoded と指定することと、合わせて POST Body に token=xoxb-xxxx&length=8192&filename=example.jpg のようなkey=value&key=value... の形式で必要な情報を記載することに注意すれば、問題なく Flux から呼び出しを行えました。
残念ながら
ここまでは、なんとかなったのですが、次に files.getUploadURLExternal API の戻り値で指定されたURLに、画像データをアップロードする部分がうまく行かずに失敗となりました。エラーとしてはこれまでと同じくFlux側の制限でエラーとなってしまったので、これ以上方法はなさそうでした。残念。
それ、Make ならできるよ!
他の iPaaS(Integration Platform as a Service)はどう対応しているのか気になったので、Make で試してみました。
Make とは?
Make は、以前は Integromat と呼ばれていた iPaaS です。同じようなサービスだと IFTTT や Zapier のほうが馴染みがあるかもしれませんね。
We’ll send HTTP POST requests to this URL when events occur. As soon as you enter a URL, we’ll send a request with a challenge parameter, and your endpoint must respond with the challenge value.
通知には Google Maps のリンクも含まれているので、クリックしてみて、ボタンをクリックした際に付加された位置情報の実際の場所を確認してみてください。
生成 AI によって推定された住所情報の精度はどうでしょうか? 私も何度か試してみましたが、推定された場所が実際の位置からかなり離れていることが多くありました。また、市単位など大まかな範囲では正確な場合もありましたが、詳細な場所については API で確認した情報よりも大きくずれていることがありました。
しかし、住所情報以外のランドマーク情報や乗換案内情報については、生成 AI が提示した場所を前提とすれば、そこまで大きくずれていないように見受けられます。この結果から、生成 AI で Reverse Geocoding を行う場合、AI とのやり取りを何度か繰り返す必要があると感じました。ワンショットで期待する情報を取得するのは難しいという印象です。プロンプトもさらに工夫が必要だと感じたので、別の機会に再チャレンジしてみたいと思います。
[mov,mp4,m4a,3gp,3g2,mj2 @ 0x11f352af0] trun track id unknown, no tfhd was found
[mov,mp4,m4a,3gp,3g2,mj2 @ 0x11f352af0] error reading header
OpenCV: Couldn't read video stream from file "_mp4_files/segment_31.mp4"Error: Could not open video file: _mp4_files/segment_31.mp4
defcreate_result_mp4(frames, frame_rate, path):
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
if frames:
height, width, _ = frames[0].shape
out = cv2.VideoWriter(path, fourcc, frame_rate, (width, height))
for frame in frames:
out.write(frame)
out.release()
else:
print("The list of frames is empty.")
print(f"A video {path} has been created.")
Webhook rate limit
Make can process up to 30 incoming webhook requests per second.
If you send more than 30 requests per second, the system returns an error with status code 429.
ref: https://www.make.com/en/help/tools/webhooks#webhook-rate-limit
$ nmap -A 192.168.8.119
Starting Nmap 7.93 ( https://nmap.org ) at 2022-12-06 15:56 JST
Nmap scan report for ATOM.lan (192.168.8.119)
Host is up (0.0059s latency).
Not shown: 998 closed tcp ports (conn-refused)
PORT STATE SERVICE VERSION
554/tcp open rtsp DoorBird video doorbell rtspd
|_rtsp-methods: OPTIONS, DESCRIBE, SETUP, TEARDOWN, PLAY, PAUSE, GET_PARAMETER, SET_PARAMETER
9999/tcp open abyss?
Service Info: Device: webcam
Service detection performed. Please report any incorrect results at https://nmap.org/submit/ .
Nmap done: 1 IP address (1 host up) scanned in 9.67 seconds
Unable to install Intel® HAXM
Your CPU does not support VT-x.
Unfortunately, your computer does not support hardware accelerated virtualization.
Here are some of your options:
1) Use a physical device for testing
2) Develop on a Windows/OSX computer with an Intel processor that supports VT-x and NX
3) Develop on a Linux computer that supports VT-x or SVM
4) Use an Android Virtual Device based on an ARM system image
(This is 10x slower than hardware accelerated virtualization)
Unfortunately, your computer does not support hardware accelerated virtualization.
Windowsの方はAndroid Studio 以外のエミュレーターも試したみましたが、ハードウェアの仮想化が必要なものが多く、ほとんどでダメでした。
ハードウェアの仮想化が必要じゃないものもありましたが、グラフィックドライバーが古くインストールできなかったり、インストールできても正常に動作しないといった感じでした。無念...