Your ideas. A week of posts. Without the daily copy-paste hour.

The free Growth Guide — by Sébastien

You finish a post. Then the distribution work starts.

Paste it into LinkedIn. Shorten it for X. Find an image for Instagram. Open Facebook, select the right page, then work through the groups. Fix the link. Check the account. Set another publishing time.

An hour disappears into moving the same idea between apps.

When you also build software, serve clients or run a local business, that hour competes with work you cannot postpone. Posting becomes something you do in bursts, followed by long gaps.

I’m Sébastien, an independent developer on Réunion Island and a father of two. My second child was born in July. During development of my tool, I pushed 205 commits in one month. There was plenty to talk about. Getting it published consistently was the missing part.

My launch campaign is now literally run by Claude through the tool’s MCP server. It drafts from my ideas, adapts posts for different networks, fills the calendar and retrieves results. I approve the content.

That currently means roughly twenty scheduled posts a week across more than six networks. Those numbers describe output. They do not establish growth or prove that a particular post sells anything.

This guide shows you how to build the publishing routine, then use the evidence it produces to improve the next week.

The system in five steps

1. Give the agent something that came from you

Start with material from your actual work: a customer question, a difficult decision, a failed experiment, a demonstration or an opinion you can explain.

“Write five posts about running a business” invites generic advice. A specific situation gives the agent something useful to shape.

Put your voice into a short file called voice.md:

# My voice

Reader: who I help and what they are trying to do.
Purpose: what I want them to understand or try.

Habits: phrases I actually use, humor, sentence rhythm.
Avoid: words I would never say and claims I will not make.

Examples: three things I wrote myself, with a note explaining
what should be preserved in each.

Approved facts: verifiable numbers, dates and experiences.
Never invent: anecdotes, customers, quotations or outcomes.

Missing information: ask me or remove the claim.

You do not need a polished brand manual. You need enough honest material for the agent to recognize your choices.

Keep updating the file when you correct a draft. Explain why a sentence felt wrong. “Too formal” helps a little; “I would describe what happened before stating the lesson” gives the agent a usable instruction.

2. Connect the agent to the calendar with specific permissions

MCP lets an assistant use functions exposed by another application. Instead of handing you text to move elsewhere, the assistant can create the draft inside your scheduler.

My setup exposes 18 tools, including plan_my_week, create_draft_post, get_analytics and list_inbox. Other schedulers may use different names.

Scopes control which operations are available: reading, writing, AI generation, media, analytics, inbox access and replies. An agent reviewing performance does not need permission to send messages.

Check what each permission actually allows. A broad write scope does not automatically enforce human approval. Configure the draft and approval workflow, then give the agent clear instructions about how to use it.

3. Give each network its own version

One network, one format. A useful starting point is to assign each version a different job:

• LinkedIn: explain a work decision and its consequences.
• X: make one specific observation that stands on its own.
• Threads: invite a conversation you genuinely want to have.
• Instagram: show something, with the necessary visual.
• A local group: address a local need and include practical details.

Treat these as editorial starting points. Your audience’s response will tell you what to adjust.

Five slightly different introductions to the same paragraph rarely make a useful distribution plan. Think about what someone gets from the post before they click anything.

4. Approve the content, then let publishing run

Before approving, check the substance: is it accurate? Would I say it? What is this post trying to do? Are the destination, link and date correct?

Once approved, the publishing step can run automatically.

In my setup, Facebook Groups, personal Facebook profiles and Skool use a browser extension where the publishing APIs in the workflow do not cover those destinations.

That route depends on the browser session and the extension’s operating requirements. Check those before relying on a scheduled delivery. Afterwards, inspect the result: a scheduled item has not necessarily been published successfully.

5. Bring the results into next week’s planning

Choose an observable purpose: visits to an article, relevant enquiries or useful conversations with potential customers.

Review whatever evidence is available against that purpose. A widely seen post may produce no meaningful action. A smaller post may start a conversation worth continuing.

Keep the questions and objections people raise. They are source material for the next batch, alongside your own notes. The system stays connected to real people while the repetitive work becomes easier.

Five recipes you can copy

These prompts use material already available in your conversation or files and ask for missing inputs. Connect your MCP scheduler to create drafts directly. Without that connection, the agent can still prepare the copy and proposed calendar for manual entry.

Recipe 1 — “Fill my week”

Context. You have scattered notes and want a manageable publishing week you can review in one sitting.

Copy this prompt:

Prepare my next Monday-to-Friday publishing week.

Read my voice file, ideas and source material available in this conversation.
If my voice, audience, objective, target networks or timezone are missing,
ask for them together in one message.

Inspect connected accounts and the existing calendar if accessible.
Use only destinations I have designated.

Find five angles grounded in my material. For each, identify its source,
the value to the reader and the purpose of the post.
If there is not enough substance, produce fewer posts and explain why.

Create no more than two network-specific versions per angle:
LinkedIn with room for explanation, X focused on one observation,
Threads conversational. Flag required assets for visual networks.

Follow my voice. Invent no experiences, numbers, testimonials or results.
Vary the openings. Do not end every post with a question.
Include a call to action only when it serves the post’s purpose.

Spread the drafts across five days without replacing existing posts.
Use configured time slots; otherwise suggest times for review.

Inspect the available MCP functions before calling them.
Use plan_my_week only if it can keep everything in draft;
otherwise use create_draft_post or the equivalent.

Do not activate publishing.
Return a table with proposed date, account, angle, full copy,
missing assets, source and draft ID. Report any failed creations.

Expected result. Up to ten drafts, each with a purpose and destination. You can approve some, rewrite others and delete anything that does not deserve to go out.

The trap. Treating the maximum as a quota. My twenty-post week reflects my own workflow. Start with a pace you can review and maintain during a busy week, including time to respond when people engage.

Recipe 2 — Website or blog → social drafts through a webhook and API

Context. Your website already produces articles or listings. You want publishing one to trigger the social preparation automatically.

A webhook is a message your website sends when an event occurs. Here, the event is a piece of content becoming public. A handler receives its details, prepares the network versions and creates drafts through your scheduler’s API.

Copy this prompt to your development agent:

In this project, implement:
website publication → webhook → social variants → scheduler drafts
→ human approval → scheduled publishing.

First identify the CMS, publication mechanism and existing integrations.
Request the scheduler’s API documentation if unavailable.
Do not invent endpoints or parameters.

Trigger on the first transition to published.
Send the stable content ID, canonical URL, title, public body,
publication date and available public media.

Authenticate the webhook using the CMS-supported mechanism.
Keep credentials server-side.
Deduplicate by source content and destination so a redelivered event
does not create another draft.

Use my voice file and chosen networks; ask for them if missing.
Generate distinct versions grounded entirely in the source material.

Create drafts only through the documented API.
Store returned draft IDs and errors.
If a response is ambiguous, check whether the draft exists before
retrying. Set a maximum number of attempts.

Verify with a test article, then deliver the same event again:
there must be one draft per destination and no public posts.
Give me the required configuration and a way to pause the workflow.

Expected result. Publishing an article starts its distribution preparation without another copy-and-paste session. You review the social versions before they go live.

The trap. Treating every edit as a new publication. Correcting a headline should not send the article around all your networks again. Make redistribution of an updated article a deliberate action.

Recipe 3 — One episode or article → a week of content

Context. You have already explored a subject in a podcast, video or article. There may be several useful posts inside it, each capable of standing alone.

Copy this prompt:

Turn the article or transcript supplied in this conversation into
five proposed posts for next week.

If you only have an inaccessible link, ask for the text or transcript.
Read my voice file. Ask for my target networks if they are unknown.

Look for five distinct angles: a concrete problem, a method,
a mistake, a qualification or objection, and a practical application.
Do not force categories the source does not support.

For each angle:
- identify the source passage or timestamp;
- explain what the reader can learn or do;
- write a standalone post for one suitable destination;
- list any required media.

Preserve the source’s meaning and credit the person behind each idea.
Use quotation marks only for exact quotations.
Do not present a guest’s experience as my own.
Flag information that may be outdated or uncertain.

Suggest a five-day progression without repeating the same conclusion
or adding a link to every post by default.
If only three angles are strong, deliver three.

Present the copy for approval.
If my MCP calendar is connected, create drafts only
and report the resulting draft IDs.

Expected result. A week that explores different parts of the subject. Someone who never reads the original article can still learn something useful from each post.

The trap. Selecting five excerpts that all make the same point. Read the batch as a single reader would. If posts two through five add nothing, extract a different lesson or publish fewer pieces.

Recipe 4 — Local announcements → Facebook Groups without gambling with your account

Context. You have a workshop, appointment opening or local service to announce. Relevant groups can help you reach people who have a reason to care.

No workflow can guarantee you will never be banned. The useful goal is to publish where your announcement is permitted and relevant.

Copy this prompt:

Prepare local distribution for the announcement in this conversation.

Ask for missing details: location, date, price, availability,
contact method, target groups and each group’s rules.
Invent no scarcity or local connection.

For each group, check accessible or supplied rules for:
geographical relevance, promotional posts, required days or threads,
external links and administrator approval.
If a rule is missing or ambiguous, mark the group “needs checking”
and exclude it from scheduling.

Write a useful version for each eligible group.
Clearly disclose my relationship to the offer.
Include practical information in the post so readers do not have
to click just to understand it.

Return: group, applicable rule, proposed copy, permitted time slot
and required approval. Create drafts only where supported.

Do not publish before I explicitly approve the copy and destinations.
Do not contact administrators on my behalf.
Suggest no filter-evasion tactics.
If a post is rejected or an account is restricted, pause that destination.
Never automatically repost rejected content.

Expected result. A reasoned selection of groups, suitable drafts and a clear list of anything you still need to verify.

The trap. Assuming different wording or longer gaps make unwanted promotion welcome. A browser extension can perform the publishing action. It does not grant permission from the community or remove platform restrictions.

Recipe 5 — This week’s results → next week’s plan

Context. Your posts have gone out. You need to decide what to continue, change and stop.

Copy this prompt:

Prepare my weekly editorial review.

Ask for my primary objective if it is unknown.
Retrieve the available analytics for the last seven days using
get_analytics or the equivalent, plus a comparable preceding period.
Use an export if direct access is unavailable.

Separate networks and formats. Show each post’s age.
Do not treat yesterday’s post and a seven-day-old post as directly
comparable without qualification.
Never invent missing metrics. “Unavailable” is not zero.

For each post, connect its intended purpose with observable results:
reach, interactions, clicks or conversions only where available.
Show the numerator and denominator for any calculated rate.

If inbox access is authorized, use list_inbox or the equivalent
to identify questions and objections. Do not reply or include
personal information in the report.

Return three decisions: continue, change, stop.
For each, cite the observations, explain the uncertainty
and propose a test that changes only one variable.

Then propose five angles for next week using these findings
and my own notes. If evidence is insufficient, suggest a baseline week.
Do not publish anything automatically.

Expected result. A few supported decisions and a fresh set of angles. Next week’s calendar has a reason to look different.

The trap. Declaring victory after one unusually successful post. Keep your cadence stable while learning. Increasing volume makes the results harder to interpret when you are already changing topics, formats and timing.

The catches you should know about

X charges for API usage. The published rate is $0.015 to create a post and $0.20 for one containing a URL—roughly thirteen times as much. Check the rates before automating significant volume. Official X pricing.

My workflow puts the link in the first reply. That is not a guaranteed billing shortcut: the reply containing the URL may itself incur the URL rate. Budget for the complete sequence.

An old Reddit account can still be filtered. Mine was twelve years old, and my first post was filtered anyway. Build in a participation period before promotion: read the community rules, contribute useful answers and build karma. Requirements differ between communities. There is no universal karma target that guarantees acceptance.

Fully AI-written content gets noticed. It happened to me twice on LinkedIn. My response is to put more human material into the process and review the intention of each post. A voice file helps, but it cannot supply an opinion you do not have. Read the draft aloud and remove anything you would not defend in conversation.

Consistency beats volume. Pick a pace that leaves room to answer people, review the evidence and do work worth sharing. A packed calendar is easy to generate. Keeping it useful requires judgment.

I run this system with PurrPlan, the tool I built. You can use any MCP scheduler that exposes the functions you need and handle the remaining steps manually. Start with one real idea, prepare a few drafts, approve them and pay attention to what comes back.

Going further

• Best time to post on X in 2026 — the slots that actually perform, with data.
• How the X algorithm works in 2026 — what Phoenix really rewards (and punishes).
• X Reach Score — score your tweet before you post it, free.
• PurrPlan vs Hootsuite — the same work, without the agency price tag.