Author: gptadmin

  • Idempotency keys: how to retry social post API calls without duplicates

    Anyone who automates social posting eventually meets this problem. A script sends a request to create a post. The connection is slow, the HTTP client gives up after thirty seconds, and the retry logic sends the request again. But the first request had already succeeded. Now the brand account shows the same post twice.

    What an idempotency key does

    An idempotency key is a value you send with a request to say “this is the same operation as before”. The GPTpost API accepts an Idempotency-Key header on create calls. If a request arrives with a key the API has already seen, it returns the result of the original request instead of creating something new.

    Without the header, each request is treated as new, so a retry creates a duplicate. That is why we recommend always sending one.

    Choosing good keys

    • Describe the logical post, not the attempt. A key like launch-2026-11-20-mastodon stays the same across retries. A timestamp does not.
    • One key per post per account, if you create separate posts per account.
    • Use only safe characters. Letters, digits and a few punctuation marks such as dots, colons and hyphens.
    • Never reuse a key for different content.

    Combine with a local record

    For longer jobs, such as scheduling a whole content calendar, keep a small local file that records which posts were created. On a re-run, the script skips what it already did, and the idempotency key covers the edge case where a request succeeded but the response was lost.

    Handle rate limits politely

    The API allows about 120 requests and 30 writes per minute per key. When you hit the limit, the response is a 429 with a Retry-After header. Wait that long, then retry with the same idempotency key.

    We use exactly this pattern to schedule our own posts. Docs and examples: GPTpost social media API.

    https://oraglegpt.org/use-cases/social-media-api/?utm_source=facebook&utm_medium=social&utm_campaign=organic90

  • How to write useful alt text in three steps

    Alt text is the written description attached to an image. Screen readers read it aloud, and it appears when an image fails to load. Most social networks let you add it, but it is often skipped or filled with something unhelpful like “image” or a file name. Writing good alt text is not hard. Three steps cover almost every post.

    Step 1: describe what is in the image

    Focus on what a sighted reader would notice first and what matters. You do not need to mention every object. “A laptop on a desk showing a calendar with posts scheduled across a week” is more useful than a list of everything visible.

    Step 2: say why it matters for this post

    The same image can need different descriptions depending on the post. A photo of your team at a conference might be described as “four people at a booth, smiling, with our logo behind them” in an event recap, and as “our support team at last year’s conference” in a post introducing the support team. Context decides what is relevant.

    Step 3: skip “image of”

    Screen readers already announce that something is an image. Starting with “image of” or “picture of” just adds noise. Go straight to the description.

    Special cases

    • Charts and graphs: write the takeaway, for example “monthly signups roughly doubled between June and September”, rather than describing each bar.
    • Screenshots of text: include the text itself, or the key part of it.
    • Decorative images: if an image adds nothing, a very short description is fine.
    • Memes and jokes: describe the joke, otherwise the point is lost.

    Make it part of the routine

    The easiest way to always write alt text is to write it at the same time as the post, not afterwards. In GPTpost, each image in a Bluesky post (up to four) has its own alt text field next to it, so the description travels with the post through approval and scheduling.

    More accessibility habits for social posts: oraglegpt.org.

  • How to cross-post a tutorial to WordPress, DEV and Lemmy without copy-paste

    Publishing a tutorial in one place is rarely enough. Your own blog keeps it under your domain, DEV puts it in front of developers who do not know you yet, and a Lemmy community can start a real discussion. Then come the short announcements on Bluesky, Mastodon and X. Done by hand, that is a lot of copy and paste.

    What goes wrong when you do it by hand

    • A typo fixed on the blog stays wrong on DEV.
    • Formatting breaks differently on each platform.
    • One of the versions gets forgotten for a week.
    • The same title is used everywhere, even when it does not fit the audience.

    A one-composer workflow

    1. Write once. Draft the full article in the GPTpost composer.
    2. Give each long-form destination its own title. WordPress, DEV and Lemmy each have a separate version. A DEV reader responds to a specific, practical title. A Lemmy community often prefers a descriptive title or a question that invites discussion.
    3. Add the short versions. A 300-character Bluesky cut, a Mastodon version with a couple of CamelCase hashtags, and an X version, each linking to the article.
    4. Approve once. All versions go through the same approval, so a reviewer sees the whole package.
    5. Schedule. Publish the long-form versions first and the short announcements a little later.

    Adapt, do not just duplicate

    Cross-posting works best when each version respects its platform. On Lemmy, add a question at the end to invite comments. On DEV, make sure code blocks are formatted properly. On your blog, link to related articles. Small changes make the content feel native everywhere.

    Know the platform rules

    Some platforms limit how often new accounts can publish, and communities have their own rules about self-promotion. Read them before scheduling a series.

    We wrote more about this here: publishing to Lemmy, DEV and WordPress from one place.

    https://oraglegpt.org/blog/cross-post-lemmy-dev-wordpress/?utm_source=facebook&utm_medium=social&utm_campaign=organic90

  • Why our comparison pages say when a competitor is the better choice

    If you read our comparison pages for Buffer, Hootsuite or Later, you will find a section titled “Choose [the other tool] if”. It lists, in plain words, situations where that product fits better than GPTpost. People occasionally ask why a company would publish that about itself. Here is our reasoning.

    Every product has gaps

    GPTpost focuses on planning, per-network versions, approvals and reliable publishing. It does not have a social inbox, it does not do social listening, and its analytics are limited to Bluesky engagement counts and short-link clicks. Some teams need exactly those missing pieces every day.

    The question is when people find out

    Someone who needs an inbox will discover that GPTpost lacks one. If they learn it on our comparison page, it is simply information. They choose a tool that suits them, and they may remember us later when their needs change.

    If they learn it on the third day of a trial, after connecting accounts and inviting colleagues, it feels like something was hidden. That person leaves with a worse impression than if they had never signed up, and they tell other people.

    Fair comparisons are also better marketing

    Readers are good at spotting one-sided comparisons. A page that says “Buffer has a free plan and already supports Mastodon and Bluesky” is more believable when it then says “GPTpost adds multi-stage approvals and client review links”. Credibility on the first claim carries over to the second.

    How we keep it fair

    • Competitor details come from their own public pages, with the date we checked.
    • “Not checked” means we did not verify something, not that the other product lacks it.
    • We remind readers that plans change and they should check the vendor’s site.

    If you spot something out of date, write to hello@oraglegpt.org and we will correct it. Read the pages: compare GPTpost with other schedulers.

    https://oraglegpt.org/compare/?utm_source=facebook&utm_medium=social&utm_campaign=organic90

  • Client review links: approvals without asking clients to log in

    Ask anyone who runs social media for clients what slows them down, and approvals will be near the top of the list. The writing gets done. The scheduling gets done. Then the posts sit in a folder, an email thread or a spreadsheet, waiting for someone who is busy with everything else.

    Why the usual methods struggle

    • Email with screenshots loses the differences between network versions and creates long reply chains.
    • Spreadsheets hide the real text in cells and drift out of date the moment you edit a post.
    • Tool logins mean another password, another invite, and often another paid seat.

    How GPTpost review links work

    A client review link opens in any browser. The client sees each post exactly as it will be published, with the separate version for each network: the 300-character Bluesky cut, the longer LinkedIn text, the Mastodon version with its hashtags. They approve or leave a comment on the draft.

    There is no account to create and no seat to pay for. Reviewers using links do not count toward your plan’s seats, so adding a client’s manager or legal contact does not change the price.

    Nothing goes out unapproved

    If a post has not been approved by the time it is scheduled, it waits. That can feel inconvenient, but it is far better than an unapproved post going live. You can combine review links with multi-stage approvals, for example an internal review first and then the client.

    Making it work in practice

    1. Send the link early in the week with a clear deadline.
    2. Keep comments on the draft itself, so feedback does not scatter across email.
    3. When a client approves late, reschedule on the calendar instead of rushing.

    You can try review links with a real client during the 14-day free trial. No card is needed and the trial never converts to a paid plan on its own. Read more: GPTpost for agencies.

    https://oraglegpt.org/use-cases/agencies/?utm_source=facebook&utm_medium=social&utm_campaign=organic90

  • How to write changelog posts that users actually read

    Release notes are usually written by the people who built the change, for the people who built the change. That is fine for the notes themselves. On social media, where your users are scrolling past dozens of other posts, it does not work.

    Lead with the user-visible change

    Compare two openings for the same release:

    • “Refactored ExportService and closed PROJ-2231.”
    • “You can now export a single month instead of your whole history.”

    The first is accurate and meaningless to anyone outside the team. The second tells a reader whether they should care. Write the user-visible change first, in plain words, and in the second person if it fits.

    A four-line structure

    1. What changed for the user. One sentence.
    2. Who it helps. “Useful if you send monthly reports to clients.”
    3. One concrete example. A short scenario, or a tiny snippet if your audience is developers.
    4. Link to the full notes. Internal names, migration steps and edge cases live there.

    Careful with numbers

    “Faster” invites the question “how much faster, measured how?” If you have a solid measurement, share it with its context. If not, describe what changed instead. Developers in particular notice vague performance claims.

    Adapt per network

    A changelog post on Mastodon can include a few CamelCase hashtags and up to 500 characters. On Bluesky you have 300. A DEV or WordPress version can explain the change in depth with its own title. Writing one core message and adapting it per network beats pasting the same text everywhere.

    Do not rely on memory

    The most common changelog post problem is that it never gets written. If your release notes have an RSS feed, GPTpost can check it on a schedule and turn each new entry into a draft. Someone still edits and approves it, but the reminder is automatic.

    More for developer relations teams: GPTpost for DevRel.

    https://oraglegpt.org/use-cases/developer-relations/?utm_source=facebook&utm_medium=social&utm_campaign=organic90

  • Short links, click counts and a weekly digest: modest analytics, clearly scoped

    Analytics is where many tools overpromise. Dashboards fill up with estimated reach, blended scores and charts for networks the tool can barely read. GPTpost takes a narrower approach, and it is worth being exact about what it measures today.

    What GPTpost measures

    • Bluesky engagement. Public counts of likes, reposts, replies and quotes for posts GPTpost published on Bluesky.
    • Short-link clicks. When you use a GPTpost short link in a post, clicks are counted, whatever network the link appeared on.
    • A weekly email digest. A summary of the above, delivered so you do not have to remember to check.

    What it does not measure

    Engagement from other networks is not read yet. There is no social inbox, no listening, and no estimated reach or impression modelling. If your role includes monthly performance reporting across many networks, a dedicated analytics suite is the better fit, and our comparison pages mention a few.

    Why clicks are a good small-team metric

    For a small team with limited time, the most useful question about a post is often simple: did it send anyone to the page we care about? Likes tell you a post was seen and liked. A click tells you someone wanted more. Clicks are also comparable across networks, which platform-specific engagement often is not.

    A 15-minute weekly routine

    1. Open the weekly digest.
    2. Note the two posts with the most clicks and the one with none.
    3. Ask what was different: topic, format, time, network.
    4. Write one adjustment for next week’s plan.

    That is enough to improve steadily without turning measurement into a second job.

    More about GPTpost: oraglegpt.org.

    https://oraglegpt.org/?utm_source=facebook&utm_medium=social&utm_campaign=organic90

  • How to schedule a social media post from Claude Code

    If you already use Claude Code for development work, you can also use it to draft and schedule social posts in GPTpost. Setup takes a few minutes. This guide walks through it step by step.

    Step 1: create an API key

    Sign in to GPTpost as a workspace owner or admin and open Settings > API & Agents. Choose Create API key, give it a name you will recognise later (for example “claude-code-laptop”), pick scopes and an expiry. For drafting and scheduling you need accounts:read, posts:read and posts:write. Copy the key straight away: it starts with gpk_ and is shown only once.

    Step 2: add the MCP server

    In your terminal, run:

    claude mcp add --transport http gptpost https://app.oraglegpt.org/mcp --header "Authorization: Bearer <YOUR_API_KEY>"

    Then start Claude Code and run /mcp. You should see the GPTpost tools, such as list_accounts, create_post and schedule_post.

    Step 3: ask for what you want

    Start with something small: “List my connected accounts, then schedule a draft for tomorrow at 9:00 announcing our new release, with a shorter version for Bluesky.” Claude will look up the accounts and create the post with a version per network.

    Step 4: check the calendar

    Open the GPTpost calendar. The post appears like any other. If your workspace requires approval, it waits for sign-off. If a version is over a network’s limit, the API rejects it with a clear error, and Claude can shorten it and try again.

    Tips

    • Use one key per machine or agent so you can revoke them separately.
    • Give keys an expiry date.
    • Keep approvals on while you get used to the workflow.

    The API and MCP server are included in every plan and the 14-day free trial, no card required. Full guide: schedule social media posts with Claude and ChatGPT.

    https://oraglegpt.org/blog/schedule-social-media-posts-with-claude-and-chatgpt/?utm_source=facebook&utm_medium=social&utm_campaign=organic90

  • Why we built an MCP server into a social media scheduler

    When we started looking at how people actually write social posts in 2026, one pattern kept showing up. The first draft often comes from an AI assistant. Someone asks Claude or ChatGPT for a launch announcement, gets something decent, then copies it into their scheduler, trims it for each network and schedules it by hand.

    The weak point in that workflow is the copy-paste step. That is where the wrong version lands on the wrong network, where a link goes missing, or where a post skips review because it felt finished.

    What MCP changes

    The Model Context Protocol lets an AI assistant call tools directly. GPTpost runs a remote MCP server at app.oraglegpt.org/mcp with a deliberately small set of tools: whoami, list_accounts, create_post, schedule_post, publish_post, upload_media, get_post_status and list_posts. An assistant can see which accounts are connected, create a post with a version per network, and put it on the calendar.

    No shortcuts for agents

    The part we spent the most time on is what an agent cannot do. It goes through the same checks as a person in the app:

    • Text over a network’s limit is rejected before anything is queued.
    • If your workspace requires approval, the post waits for approval.
    • Plan limits apply in the same way.
    • An API key never has more rights than the person who created it, and it stops working if that person loses their role.

    That makes it reasonable to let an agent draft and schedule, while people keep the final decision.

    Where it works today

    Claude Code connects with one claude mcp add command and a workspace key. Claude Desktop works through mcp-remote. Custom connectors on claude.ai need OAuth, which the endpoint does not support yet, so we say so plainly. ChatGPT can use the OpenAPI schema through a Custom GPT action.

    Details and setup: GPTpost MCP server for AI agents.

    https://oraglegpt.org/use-cases/mcp-server-for-ai-agents/?utm_source=facebook&utm_medium=social&utm_campaign=organic90

  • Dogfooding: scheduling our own 90-day calendar through the GPTpost API

    The posts you will see on GPTpost’s own accounts over the next three months were not entered by hand. We wrote a content calendar, and a short script scheduled it through the same agent API that every GPTpost plan includes. Here is how that works and what we learned setting it up.

    The calendar file

    Each entry in the calendar has a slot identifier, a time in UTC, and a version of the post for every network we publish to: a short cut for Bluesky, a slightly longer one for Mastodon, a titled article for WordPress and Lemmy, and so on. Writing the versions up front means the script never has to trim text, which is where automated posting usually goes wrong.

    The script

    1. List the connected accounts with GET /api/v1/agent/accounts and keep only the ones meant for brand posts.
    2. For each calendar slot and account, create and schedule the post with POST /api/v1/agent/posts, including schedule_at.
    3. Send an Idempotency-Key built from the slot and the account, and record the result in a local state file.
    4. Pace the writes to stay under the documented limit of about 30 writes per minute, and back off when a 429 arrives with Retry-After.

    Why idempotency matters here

    A long run will hit a timeout or a rate limit at some point. Without an idempotency key, the safe-looking fix (run it again) creates duplicates. With one, re-running is boring: posts that exist are skipped, and new accounts are filled in. That also lets us connect another network later and simply run the same script again.

    Same rules as people

    The API does not bypass anything. Text over a network’s limit gets a 422 before anything is queued, approval workflows return a 409, and plan limits a 402. For our own account that was a useful check on the calendar itself.

    Try it yourself

    The API and the MCP server are on every plan and the 14-day trial. The docs and OpenAPI schema are public, so you can judge them before signing up. Start here: social media API for developers and AI agents.

    https://oraglegpt.org/use-cases/social-media-api/?utm_source=facebook&utm_medium=social&utm_campaign=organic90