Prompts and tool calls

You ask in plain words and the agent makes the calls. This page shows prompts that work and the calls behind them, so you can tell what went wrong when a result looks off. The examples use Bluesky, WordPress and DEV. The analytics example uses Instagram. Most requests follow the same order: find the channel with integrationList, read its rules with integrationSchema, look up any value a setting needs with triggerTool, bring in media, then create the post with integrationSchedulePostTool. Every tool is on MCP tools.

Name the channel, the day and the time, and say whether you want a draft. “Draft” keeps the post on your calendar until you schedule it yourself.

Schedule a post

Prompt

Schedule a Bluesky post for Thursday October 1 at 9:00 UTC: "Our spring menu is live. Come by this weekend."

The tool calls

The agent finds the Bluesky channel, reads its rules, then creates the post.

integrationList

{}

integrationSchema

{ "platform": "bluesky", "isPremium": false }

Bluesky answers with a 300-character limit and No additional settings required, so settings stays empty:

integrationSchedulePostTool

{
  "socialPost": [
    {
      "integrationId": "your-bluesky-channel-id",
      "isPremium": false,
      "date": "2026-10-01T09:00:00Z",
      "type": "schedule",
      "shortLink": false,
      "postsAndComments": [
        { "content": "<p>Our spring menu is live. Come by this weekend.</p>", "attachments": [] }
      ],
      "settings": []
    }
  ]
}

It returns the new post’s postId.

Post a thread

Prompt

Post this as a Bluesky thread on Friday October 2 at 10:00 UTC, with the second part five minutes after the first.

The tool calls

Each item after the first in postsAndComments becomes a reply in the thread. delay is the wait in minutes after the item before it.

integrationSchedulePostTool

{
  "socialPost": [
    {
      "integrationId": "your-bluesky-channel-id",
      "isPremium": false,
      "date": "2026-10-02T10:00:00Z",
      "type": "schedule",
      "shortLink": false,
      "postsAndComments": [
        { "content": "<p>Three things we changed in the spring menu.</p>", "attachments": [] },
        { "content": "<p>First: every dish now has a smaller size.</p>", "attachments": [], "delay": 5 }
      ],
      "settings": []
    }
  ]
}

On some networks the extra items become comments instead, and on others only the first item is published: see comments and threads.

Draft a WordPress article

Prompt

Write a short WordPress article about our spring menu and save it as a draft for Friday at 8:00 UTC.

The tool calls

integrationSchema for wordpress says a post needs a title and a type, the WordPress post type, and lists the helpers that look values up. The agent asks the site for its post types:

triggerTool

{ "integrationId": "your-wordpress-channel-id", "methodName": "postTypes", "dataSchema": [] }

Answer

{ "output": [ { "id": "posts", "name": "Posts" }, { "id": "pages", "name": "Pages" } ] }

Then it saves the draft with the id it picked:

integrationSchedulePostTool

{
  "socialPost": [
    {
      "integrationId": "your-wordpress-channel-id",
      "isPremium": false,
      "date": "2026-10-02T08:00:00Z",
      "type": "draft",
      "shortLink": false,
      "postsAndComments": [
        { "content": "<h2>What is new</h2><p>Every dish now comes in a smaller size.</p>", "attachments": [] }
      ],
      "settings": [
        { "key": "title", "value": "Our spring menu" },
        { "key": "type", "value": "posts" }
      ]
    }
  ]
}

categoriesList and tagsList work the same way. Their IDs go into the categories and tags settings as numbers.

Publish a DEV article with tags

Prompt

Publish my release notes on DEV now, titled "Version 2.4 is out", tagged webdev and javascript.

The tool calls

DEV needs a title and takes up to four tags. The agent looks the tags up:

triggerTool

{ "integrationId": "your-dev-channel-id", "methodName": "tags", "dataSchema": [] }

Each tag comes back as a value and a label, and goes into settings the same way:

integrationSchedulePostTool

{
  "socialPost": [
    {
      "integrationId": "your-dev-channel-id",
      "isPremium": false,
      "date": "2026-09-23T12:00:00Z",
      "type": "now",
      "shortLink": false,
      "postsAndComments": [
        { "content": "<p>Version 2.4 adds offline mode and fixes the login loop.</p>", "attachments": [] }
      ],
      "settings": [
        { "key": "title", "value": "Version 2.4 is out" },
        { "key": "tags", "value": [ { "value": 1, "label": "webdev" }, { "value": 2, "label": "javascript" } ] }
      ]
    }
  ]
}

The tag numbers here are placeholders: the agent uses the ones tags returned.

Add an image

A post takes media from your PostQueen media library, never a link to another site or a file on your computer. There are three ways in.

From a link
From your device
Generate one

Prompt

Use the photo at https://example.com/spring-menu.jpg for Thursday's Bluesky post.

The agent calls uploadFromUrlTool with the link. It returns the file’s id, path and status, and the path goes into attachments.

Prompt

I want to add a photo from my laptop to this post.

In Claude, Claude Cowork and ChatGPT the agent calls uploadWidgetTool, and an Upload Media panel opens in the chat. Add the file there. The agent then calls uploadWidgetStatusTool and uses the path once the file is ready. Other apps show no panel: give the agent a public link instead.

Prompt

Make a square image of a bright spring salad on a wooden table for this post.

The agent calls generateImageTool with a prompt and "orientation": "square". It is paid from the workspace’s credits and returns the image’s path.

A video comes back as processing while PostQueen prepares it, and its path can change when it is done. From the upload panel, the agent waits until uploadWidgetStatusTool says ready and uses that path. A video from a link has no status tool yet, so check that post on your calendar before it goes out.

Make a video for a post

Prompt

Make a short vertical video of our spring salad for Thursday's Bluesky post.

The tool calls

The agent lists the generators first, to learn their identifiers and inputs:

generateVideoOptions

{}

Then it starts the video. seedance makes one 8-second clip with sound; images stays empty for a video from the prompt alone:

generateVideoTool

{
  "identifier": "seedance",
  "output": "vertical",
  "customParams": [
    { "key": "prompt", "value": "Slow push-in on a bowl of spring salad on a wooden table by a sunny window, warm morning light." },
    { "key": "images", "value": [] }
  ]
}

It returns a jobId at once. Making the video takes minutes, so the agent asks every 30 seconds or so:

videoStatusTool

{ "jobId": "the-job-id" }

Answer

{ "status": "completed", "id": "the-media-id", "url": "https://cdn.example.com/k3l4m5n6o7.mp4" }

The url goes into attachments of integrationSchedulePostTool. The video is paid from the workspace’s credits, and it is also in your media library.For a narrated slideshow, the agent uses image-text-slides. It gets a voice first with videoFunctionTool ("identifier": "image-text-slides", "functionName": "loadVoices") and passes its id as voice, with a prompt. During the free trial Seedance is locked, so the agent gets This video is not available in trial mode and can use image-text-slides instead.

Check and change what is scheduled

Prompt

What do I have scheduled next week? Put the WordPress draft up for review instead.

The tool calls

postsListTool lists the posts in the window:

postsListTool

{ "startDate": "2026-09-28T00:00:00Z", "endDate": "2026-10-05T00:00:00Z" }

postReadTool reads the draft in full, then postSettingsTool changes one setting. WordPress’s status is the state the article gets on your WordPress site when PostQueen publishes it: publish, draft, pending or private. In PostQueen the post stays a draft until you schedule it.

postSettingsTool

{ "id": "the-post-id", "settings": [ { "key": "status", "value": "pending" } ] }

Only the settings you pass change. The text and the time stay as they are.

An agent cannot delete a post, change its text or time, or move it between draft and scheduled over MCP. Do that on the calendar, or with the CLI or the API.

Ask how posts did

Prompt

Which of my Instagram posts did best this month, and how many impressions did my posts get in the last 90 days?

Bluesky, WordPress and DEV report no analytics, so this example uses Instagram.

The tool calls

analyticsPostsTool

{ "date": 30, "platform": "instagram", "sort": "reactions", "limit": 1 }

analyticsSummaryTool

{ "date": 90 }

The window picks posts by publish date, and the numbers are each post’s totals so far. null means the network does not report that number. The networks that report numbers are listed on MCP tools.

Hand over the whole request

Prompt

Ask PostQueen to plan three Bluesky posts about our spring menu, one a day from Monday at 9:00 UTC.

On the key address and the Bearer header, the agent can pass the request to ask_postqueen, PostQueen’s own agent, which runs the other tools itself. It writes out the drafts and waits for your yes before it schedules them. The sign-in address has no ask_postqueen: there the agent makes the calls itself.

Next steps

Every agent

Which agents can post to PostQueen, and how each one connects.

The MCP tools

What each of the 22 tools does, and what an agent cannot do.

The order an agent calls the tools in, with prompts that work.

Troubleshooting agents

What to do when a connection or a command fails.

Start here

Assistants and bots

Coding agents

Editors and more

MCP reference

CLI reference

Troubleshooting

Prompts and tool calls

Schedule a post

Post a thread

Draft a WordPress article

Publish a DEV article with tags

Add an image

Make a video for a post

Check and change what is scheduled

Ask how posts did

Hand over the whole request

Next steps

Every agent

The MCP tools