Example agent prompts and the tool calls they trigger
Each prompt below works in any MCP client connected to Socifyr (Claude, Cursor, ChatGPT, Codex). The tool calls show what a well-behaved agent sends. Models vary, so your agent may take a slightly different path. Slugs like x-myhandle are examples; the agent finds yours with list_connections.
How do I get an agent to post to several accounts?
Prompt: "Post 'Shipped v2.0 today' to my LinkedIn and X accounts."
{ "tool": "list_connections", "arguments": {} }
{
"tool": "post_text",
"arguments": { "caption": "Shipped v2.0 today", "connections": ["linkedin-myhandle", "x-myhandle"] }
}Result text: Post submitted successfully! ID: <id> Request ID: <requestId>. Tools: post_text.
How do I get an agent to schedule a week of posts?
Prompt: "Write five short LinkedIn posts about our new queue feature and schedule one per weekday at 9am UTC starting 2026-12-07."
{
"tool": "schedule_posts",
"arguments": {
"posts": [
{ "text": "Monday: how the queue retries failed channels", "connections": ["linkedin-myhandle"], "scheduledAt": "2026-12-07T09:00:00Z" },
{ "text": "Tuesday: idempotency keys and safe retries", "connections": ["linkedin-myhandle"], "scheduledAt": "2026-12-08T09:00:00Z" }
]
}
}Result: Scheduled 5 of 5 posts. plus a line per failed row. Needs a plan with scheduling. See Bulk schedule.
How do I get an agent to post an image from my library?
Prompt: "Find banner.jpg in my media library and post it to Instagram with the caption 'Launch day'."
{ "tool": "list_media", "arguments": { "type": "image", "search": "banner" } }
{
"tool": "post_image",
"arguments": { "connections": ["instagram-mybrand"], "images": ["6a04f715d3d7d93575e65d13"], "text": "Launch day" }
}How do I get an agent to check a post and retry failures?
Prompt: "Did my last post go out everywhere? Retry anything that failed."
{ "tool": "list_posts", "arguments": { "limit": 1 } }
{ "tool": "check_post_status", "arguments": { "postId": "6a05032a84584482ace13d5f" } }
{ "tool": "retry_post", "arguments": { "postId": "6a05032a84584482ace13d5f" } }If a channel failed with reconnect_required, a good agent fixes the account first, as in the next prompt. See Check status and retry.
How do I get an agent to fix an expired account?
Prompt: "My X post failed with an expired token. Fix the connection and retry."
{ "tool": "list_connections", "arguments": {} }
{ "tool": "refresh_connection", "arguments": { "connection": "x-myhandle" } }If refresh fails, it calls reconnect_connection, gives you the URL, waits for you to approve it, then calls retry_post:
{ "tool": "reconnect_connection", "arguments": { "connection": "x-myhandle" } }Open the link yourself and approve the same account. Both tools need an unrestricted key. See Connect, refresh and reconnect.
How do I get an agent to report follower growth?
Prompt: "Which of my accounts grew the most in the last three months?"
{ "tool": "get_follower_growth", "arguments": { "period": "last_3months" } }
{ "tool": "compare_accounts", "arguments": { "period": "last_3months" } }LinkedIn and TikTok report no follower count, so an agent should say so instead of guessing. See Analytics.
How do I get an agent to draft copy and wait for my approval?
Prompt: "Write a LinkedIn and X caption for the launch, show me both, and only post after I say go."
{
"tool": "generate_caption",
"arguments": { "description": "We launched bulk scheduling from CSV", "platforms": ["linkedin", "x"], "tone": "professional" }
}After you reply "go", it calls post_text for each platform's caption. generate_caption needs an AI key on the workspace. See AI captions.
How do I write prompts that work well?
- Name the platforms and accounts, or tell the agent to list connections first.
- Give times in UTC or with an offset. The tools take ISO 8601 strings.
- Ask for a confirmation step before anything that publishes.
- Use a key with required approval when the agent runs unattended.
FAQ
Can the agent edit a post it already scheduled?
Yes, with update_post while it is a draft or an unpublished scheduled post. A connections value replaces the full set of accounts.
Why did the agent say a tool is not allowed?
The key has guardrails. See Approvals and guardrails.