Help center

Connect your AI: Claude, ChatGPT and other assistants working with your companies

ōrbita has an MCP connector so your own AI can work with your companies. In Claude you add it from its connector directory; in ChatGPT or another compatible assistant you paste the address https://mcp.orbitasolutions.org/mcp. Then you sign in to ōrbita and choose what you allow. From then on it reads the context, posts and metrics of your companies and leaves drafts for you to review. It acts with your own permissions and, if you grant it that permission and ask for it and confirm it in the chat, it also schedules a post or replies to a comment for you. Included in Satellite, Star and Galaxy.

Updated Author: Mónica Lorenzo In the app: /me/agents

What it is

If you already pay for an AI assistant (Claude, ChatGPT or another), you can connect it to ōrbita and ask for things in its own chat: “rewrite the copy of Thursday’s post in a warmer tone”, “prepare three drafts for the autumn campaign”, “explain this month’s metrics”. Your assistant reads what ōrbita knows about your company (strategy, brand voice, catalog, reviews, calendar, metrics) and writes the result straight into ōrbita.

The connection uses MCP, the standard AI assistants use to connect to other applications. There are no keys to copy and nothing to install: in Claude you add it from its connector directory and in other assistants you paste an address; then you sign in to your ōrbita account and authorize.

The connector address is always the same:

https://mcp.orbitasolutions.org/mcp

You will also find it, with a copy button, in My account > My agents.

What your AI can and cannot do

Your AI acts with your identity and your own permissions: it can never do more than you can do in each company. Whatever your level in a company, your AI can do exactly what you can do there, and nothing more.

When you connect, ōrbita shows you a screen with what your AI will be able to do:

  • Read: your companies and their context, the catalog, social proof, posts and the calendar, key dates, the brand kit (colours, fonts, hashtags, notes, which logos you uploaded and your links to brand material), the metrics of what has been published, your Google listing (details, description, hours, attributes, services and your average rating; it is a read permission that comes with the basic ones) and the open tasks of the company (the same ones you see in Tasks), with their comments and attached files.

  • Optional permissions, each with its own checkbox:

    • Content: create drafts and edit posts, duplicate them (also into another of your companies) or move them to another company, add images or videos from a public link and change who they are assigned to (the new person gets an email and a notification). With this permission alone it does not publish or schedule: you review the drafts in ōrbita. It can take a scheduled post back to draft while it has not gone out yet.
    • Schedule or publish a post on your networks, only when you ask for it and confirm the text, the networks and the date in the chat. One yes can cover several posts if you see them together in the chat, each with its text, networks and date; each one is sent separately. TikTok stays out: its settings are yours to choose in ōrbita. ōrbita keeps what your assistant scheduled (text, networks and date) for 90 days, never your conversation. Details in “Schedule, publish and reply from your AI”, below.
    • Plan: edit the company profile and the sections of the communication plan.
    • Tasks: create tasks for you or for your team and requests for your AI, move them to in progress, attach files to them from a public link and mark them as done with a note of what it did. As in the app, it can close any open task of the company, but it asks you before closing one.
    • Catalog: create, edit, pause and delete products and services, decide which ones show in the marketplace and put photos or videos on them from a public link.
    • Social proof: create, edit and delete your reviews and social proof (only in the companies where you are an administrator).
    • Key dates: create, edit and delete calendar dates.
    • Brand kit: edit colours, fonts, hashtags, notes and links to brand material (only in the companies where you are an administrator). Logos are uploaded by you in ōrbita.
    • Read the interactions inbox of your networks: comments, mentions and Google reviews. Your assistant reads them, looks up the saved replies and drafts replies for you; to send them from the chat it also needs the reply permission, or you send them in ōrbita.
    • Reply to comments, mentions and Google reviews on behalf of your company, only when you ask for it and confirm the exact text in the chat. One yes can cover several replies if you see them together in the chat, each with its text; each one is sent separately. On Google, a new reply replaces the previous one. Your yes in the chat also counts for reviews below 4 stars. Private messages stay out. With this permission it also creates, changes, deletes and comments on the saved replies of the inbox, as you would in the app.
    • Change your Google listing: modify the description, the phones, the website, the hours and the products and services from your catalog that appear on it (the ones you mark with “Show as a service on your Google Business Profile”), only when you ask for it. Your listing is public and the change is seen on Google: your assistant shows you the exact change and only makes it when you say yes in the chat. You change the name, the categories and the address yourself in ōrbita, and you need to be an administrator of the company. Details in Your Google listing in ōrbita.
    • Save metrics reports in ōrbita: your agent leaves in Metrics, Reports, the report it has written with your data. It never shares it: you create the public link. Details in Metrics reports with your agent.
    • Read marketplace reviews.
    • Reply to marketplace reviews on behalf of your company (you need to be an administrator).

    They come ticked; untick whatever you do not want to allow. Whatever you untick you can give later in My account > My agents with the Give permission button, without connecting it again.

What it can never do, whatever permissions it has:

  • Publish, schedule or reply without you asking for it and confirming it in the chat, even with the permission: without your confirmation it leaves drafts. On TikTok it never schedules.
  • Send private messages or automatic replies.
  • Tick for you the declarations each network requires to be made by hand (for example, branded content on TikTok or “made for kids” on YouTube).
  • Generate images or video. It can put media on a post or a catalog product, but only what is already attached to one of your tasks or what arrives from a public link you give it (see “Images and videos” below).
  • See your billing or manage the members of your companies.

What your AI creates arrives as a draft, and you find it where you always do: without a date, in Drafts; with a date, in the Calendar. Your AI gives you the direct link to every post it touches. Every action it takes is signed in ōrbita with your agent’s name and yours, for example “ChatGPT de {your name}” (never “AI”), and, if it goes outside ōrbita, with how you approved it: with your yes in the chat or with the ōrbita button. You see it that way in the history of every task, post, plan, catalog item, key date or Google listing, in comments, on the post’s stamp and on “Written by” for key dates. In emails and assignment or comment notices it appears as “El asistente de {your name}”. You can edit or delete any comment of yours in ōrbita, including the ones your assistant leaves.

Watch out for one common confusion: a date on a post is a suggestion until someone schedules it. Your AI putting 30 September on it only means the post is waiting for you on that day in the calendar; it is not queued on any network and it will not go out by itself. You schedule it in ōrbita, or you ask your assistant to, with the scheduling permission and your confirmation in the chat (next section). If your assistant tells you something “is scheduled” without having asked you to confirm, ask it for the real state of the post: the connector returns it in every answer.

The full list of what the connector offers your assistant, tool by tool, is in the tool reference.

Schedule, publish and reply from your AI

If you tick the permissions Schedule or publish a post and Reply to comments, mentions and Google reviews when connecting, your assistant can finish the job without you leaving the chat. The rules are the same in Claude and in ChatGPT:

  • Only if you ask for it. Your assistant never schedules or replies on its own initiative: it drafts, shows you and waits.
  • You confirm the exact text, the network and the date. Before acting it shows you the post as it will go out (or the reply as it will be sent), on which networks and when (or “now”), and only acts once you say yes. It cannot change anything after your confirmation: if you want another text or another date, you ask for the change and it shows it to you again.
  • Several with one yes, if you have seen them. One yes can cover several posts or replies when your assistant shows them to you together in the chat, each with its exact text, networks and date. Each one is sent separately.
  • The same rules of each network. If the post goes to YouTube, the connector returns to your assistant the same notice you see in ōrbita (you certify that the video complies with the YouTube Terms of Service) so that it shows it to you before scheduling. What was already scheduled or published is not touched: to change it, take it back to draft first.
  • TikTok stays as a draft. TikTok requires you to choose the privacy level yourself, see the preview with the account name and tick music and the commercial-content or AI disclosures. If you ask your assistant to schedule a post with TikTok, it does not schedule it: it explains that TikTok is the one preventing it, lists those fields and gives you the direct link to the post in ōrbita so you finish it there (or offers to take TikTok off the post and schedule the rest).
  • Replies on Google: a new reply to a review replaces the previous one. Private messages: stay out.
  • Google listing and reviews below 4 stars: also with your yes. A change on your Google listing or on the catalog services that appear on it, or a reply to a review below 4 stars, counts with your yes in the chat, like the rest. The ōrbita button is only needed if the company owner has turned on Only in ōrbita in its settings: then your assistant gives you the link to approve it there.
  • What ōrbita records. For every action, ōrbita keeps who asked for it, with which assistant, the exact content with its fingerprint and how it was confirmed: in the chat, as your assistant declares, or with the ōrbita button. Never your conversation or your words. You can see it for 12 months; after that it is blocked, only to handle claims, and deleted after 4 years. You are responsible for what your assistant does with your connection as if you had done it yourself: only confirm what you have reviewed. See the privacy policy and the terms.

If you do not tick those permissions, everything stays as before: your assistant leaves drafts and you schedule and reply in ōrbita.

Set up a new company with your AI

Your assistant can also set up with you a company you have just created, step by step and in one conversation. Ask it like this: “Help me set up {company} in ōrbita, step by step: start with the company guide and follow its script”. Before saving anything it confirms the company with you and tells you which sources it has about it (what you tell it, what you attach, your website or other data within its reach). It shows you what it finds with its source, asks you only for what is missing and does not use data from another company or guesses. At every step (website, questions about the business, catalog, key dates, customer questions) it gives you its link to ōrbita and waits for you to say “done” or “skip”. You always confirm the plan: in ōrbita, with the “Confirm plan” button in Brand > Strategy, or in the chat, where your assistant shows it to you, applies the changes you ask for and only confirms it when you tell it to. With the plan confirmed it can request the first month of content. You connect the networks in Company settings > Connections, and you create the company (it counts towards your plan). If your AI lacks a permission for a step, it tells you which one and gives you the link to give it.

Images and videos

With the content permission or the catalog permission, your AI can put on a post or a product an image or a video you give it by a public link (Google Drive, Canva, your own website) or that is already attached to one of your tasks. It does not generate images or video, and if the link is private, temporary, or a file you passed it through its own chat, it explains why and points you to uploading it by hand. Full detail, limits and an example phrase are in Upload images and videos from your AI.

Connect Claude

ōrbita is in the Claude connector directory, so there is no address to copy. It works on every Claude plan.

Connect it from theClaude directory

  1. Open the ōrbita listing in the Claude directory and press the button to connect it. You can also find it inside Claude (web or desktop app), in Customize > Connectors, by searching the directory for ōrbita.
  2. ōrbita opens: sign in to your account if you were not already, review the permissions and press Authorize.
  3. In a chat, press + > Connectors and turn ōrbita on. Try: “list my ōrbita companies”.

On the listing you will see the Community label and a notice from Claude: it is the general text Claude shows on connectors published by other companies, which pass its automated review to enter the directory but are not verified by Anthropic itself. The connector’s developer is ōrbita solutions and the address it connects to is always https://mcp.orbitasolutions.org/mcp.

Once connected on the web or desktop, the connector also appears in the Claude mobile app. From Customize > Connectors you can decide, tool by tool, whether Claude acts without asking or asks for permission every time; for the ones that write, we recommend that it asks.

If you prefer to add it by hand, you can create it as a custom connector: in Customize > Connectors, press +, choose Add custom connector, give it a name (for example, ōrbita), paste the connector address and press Add. If Claude asks how to sign in, leave the default options. On the free plan Claude allows a single custom connector, so there the directory is the better route.

If you use Claude on a Team or Enterprise plan, whoever administers the organisation has to add the connector first; then each person connects it with their own ōrbita account. The official steps are in the Claude documentation on directory connectors and on custom connectors.

Connect ChatGPT

In ChatGPT, custom connectors are added from developer mode, a beta feature from OpenAI that is only available on the web (once connected, the connector also answers in the mobile app). OpenAI often changes the names and paths of these menus, and decides which plans can use developer mode and whether the connector can write or only read. These steps are the ones we saw in September 2026, in the Spanish interface, so the menu names below are descriptions rather than exact labels; if your screen does not match, the reference is the official OpenAI article.

  1. Turn on developer mode. In ChatGPT on the web, click your profile picture, open the settings, go to the plugins section, open developer mode and turn it on.
  2. Create the app. Go back to the plugins section, click + at the top right and choose to create an app, then to create an MCP app. Do not choose to create or upload a plugin.
  3. Fill in the whole screen in one go:
    • Icon: not needed.
    • Name: ōrbita
    • Description: Create and plan the social media content of your companies
    • Connection > Server URL: https://mcp.orbitasolutions.org/mcp
    • Authentication: OAuth. Leave the advanced OAuth settings as they are.
    • Tick the box confirming you understand and want to continue, then click create.
  4. Authorize in ōrbita. ōrbita opens: sign in if you were not already, review the permissions and click Authorize. Back in ChatGPT, ōrbita shows up in the plugins section of the settings.

Done, ChatGPT is now connected to ōrbita! Open a new chat, turn ōrbita on in that chat and try things like these:

  • “List the companies I have in ōrbita.”
  • “Interpret this month’s social media statistics and tell me what I should take from them.”
  • “Which social networks do I have connected in ōrbita?”
  • “Leave me a draft for next week about my main product.”
  • “Look at what I have published lately and suggest ideas for my next posts.”

Whatever ChatGPT leaves arrives in ōrbita as a draft, except what you asked it to schedule or reply and confirmed with the permissions above.

One detail: ChatGPT stores the connector’s list of functions as it was on the day you added it. When ōrbita adds new functions, delete the connector in ChatGPT and create it again to see them.

Connect Claude Code and Cursor

Claude Code. In the terminal:

claude mcp add --transport http orbita https://mcp.orbitasolutions.org/mcp

Then, inside Claude Code, type /mcp, choose ōrbita and follow the sign-in that opens in the browser.

Cursor. Add the server in ~/.cursor/mcp.json (for all your projects) or in .cursor/mcp.json inside a project:

{ "mcpServers": { "orbita": { "url": "https://mcp.orbitasolutions.org/mcp" } } }

Cursor opens the ōrbita sign-in the first time it uses the server.

Any other client compatible with MCP over HTTP and OAuth sign-in connects the same way: all it needs is the address.

Your tasks and your requests

Your AI sees and can work on the same open tasks you see in Tasks for each company, with their description, comments and attached files: yours and your team’s, including requests for another person’s AI and the ones the communication plan generates. The higher your access level in the company, the more tasks you see and the further your AI can reach.

Since not every task of yours is meant for it, before starting on an ordinary task of yours it asks you, one at a time, whether it should do it. If you say yes, it does it; if you say no, it leaves it as it is.

If you want it to do something without asking you, leave it as a request with Ask my agent (Ask my AI if your app has not updated yet):

  • From a post: you say what you want (for example, “write the copy in the usual tone”) and the request stays linked to that post.
  • From a calendar day: ōrbita creates an empty post for that day and the request for your AI to write it.

You can ask your own AI or that of another person in the company; the request is picked up by that person’s AI, with their permissions. Requests appear in Tasks with the “For your agent” label (or “For your AI”).

Your AI can also create tasks for you or for someone on your team: ask it, for example, “create a task for Friday: ask the client for product photos” and it appears in Tasks, with its date. If you do not tell it who for, it assigns it to whoever authorised it; give it the email or full name of another person in the company and it assigns it to them. Creating it does not do it or close it: the task stays open for whoever receives it. A request for your AI can only be for yourself: if you tell it the task is for your AI, it is created as a request, with the “For your agent” label, and your AI will do it in another conversation without asking you.

If the task has an image or a video attached, your AI can put it on the post. It copies it inside ōrbita, so the file never leaves ōrbita and the task keeps its own. The same limits apply as when you upload it yourself: images up to 10 MB and videos up to 100 MB, with at most 20 images and 1 video per post. It is added after the ones the post already has; it replaces none.

When your AI does any work on a task (for example, creating the post it asks for), it moves it to in progress without asking you. When it is done, it asks whether to mark it as completed or leave it in progress. If that work is creating a post from the task, the post stays linked to it.

Important: ōrbita only notifies your agent if you give it an email inbox. With an inbox, when it is assigned a task or mentioned, a notice with no data reaches that address; if it hears about something that way and you are not in the chat, it only works inside ōrbita (drafts, comments, tasks and plan sections), and anything seen outside ōrbita or that costs money waits for your yes in the chat. Without an inbox, assistants work the other way round: they are the ones who ask. Either way, your agent looks at your tasks when you tell it to or when it has a scheduled task that does it on its own. How to give it an inbox: Connect an agent. Ask for the whole thing: “check my pending tasks in ōrbita, do the requests and mark them as done”. If you do not say so, some assistants do the work and leave the task open. When it closes one, your AI leaves a note with what it did.

Scheduled task: picking up requests on its own

If your assistant has scheduled tasks, you can have it check your requests every day without being asked. In Claude they live in the sidebar, under Scheduled > New task: choose the frequency and write the standing instruction. If you cannot see that option, your plan does not include it.

The wording matters more than usual here. A scheduled task starts in a fresh session, with nothing you said before, and a generic instruction ends in “I found no requests” without having looked. The key is to name the connector and name the tools. Since you are not there to answer during a scheduled task, this text has it do only your requests and leave you the list of your other tasks:

Use the ōrbita connector (MCP) I have connected. First call list_companies; then list_pending_requests for each company. For each task with addressed_to_ai true: call start_request, then get_request_brief, do what it asks on the linked post with update_proposal_caption or update_proposal, put on it with attach_request_media the files of the task it asks for, and close it with complete_request and a short note. Leave the tasks with addressed_to_ai false untouched: at the end give me their list so I can tell you which ones to do. Do not touch anything that is not in a task. If any call to the connector fails, give me the literal error instead of summarising it.

If you would rather not fight with the settings, ask your own AI to walk you through it. Paste this into its chat:

I have the ōrbita MCP connector connected (https://mcp.orbitasolutions.org/mcp) and I want to create a recurring task in you that picks up my requests every day. Walk me through your interface step by step, ask me for the frequency and the time, and at the end give me the text of the task ready to paste. The official steps are at https://orbitasolutions.org/en/help/connect-your-ai/: do not invent any.

Claude’s own steps are in its help. Everything the task does reaches ōrbita as a draft: since you are not there to confirm during a scheduled task, it cannot schedule or reply, and you still review and schedule yourself. Two normal things you will see: the task may ask you for permission to use the connector inside the chat even though you already authorised it, and if it says it left something “scheduled”, check in ōrbita: what it leaves are drafts with a suggested date.

Control and withdraw access

In My account > My agents you see every AI you have connected, when you connected it, when it was last used and which permissions it has. From there you can:

  • Remove a specific permission without disconnecting it.
  • Give a permission it does not have, with the Give permission button, without connecting it again.
  • Disconnect the AI altogether. What it had already created in ōrbita is left untouched.

Removing the connector inside Claude or ChatGPT does not withdraw access in ōrbita: always do it from that section. And every time you add the connector in your assistant a new connection is created; the ones that stop being used are removed on their own after a while.

If you manage a company with more people, in Company settings > Members and agents you see which people have an agent with access to that company and you can remove the access of someone’s agents to it. The agents stay connected to that person’s account and the person keeps their access to ōrbita. It applies to all their agents: even if they disconnect and reconnect them, they will still have no access until someone unblocks them from that screen.

To give the access back you need to be whoever removed it, the owner, or at a higher level.

Notices from your agents. An agent cannot remove or give back another agent’s access: only a person can, in ōrbita. If it sees something odd, or you ask it to, it leaves a notice. The people who can act get an email and, in Company settings > Members and agents, see the notice with two buttons: remove the access or dismiss it. No permission changes until someone confirms it. An agent wrote the text and it may be wrong or may have been manipulated, so check before you act.

Automatic pause. If an agent makes many changes, or deletes or publishes many things, within a few minutes, ōrbita pauses it on that company. It keeps working on the others. The people who can act get an email and, in Company settings > Members and agents, see the notice with two buttons: Resume or remove the access. The pause does not expire: it lasts until someone decides. The person the agent belongs to and those who can manage the agents of that company see it.

If something does not work

  • Your AI says it has no permission for something. Either you do not have that level in that company, or that AI does not have that permission. In My account > My agents you see the permissions it does not have, each with the Give permission button: you give it without connecting it again. Your AI tells you which one it lacks and gives you the link.
  • Your agent cannot change the Google listing or save the metrics report. It lacks the permission “Change your Google listing” or “Save metrics reports in ōrbita” (for example, if you connected it before they existed). In My account > My agents, under “Permissions it does not have”, press Give permission. In ChatGPT, also remove the connector and create it again (see “Connect ChatGPT”) so that it sees the new tools.
  • Your AI does not see a company. Check that you have access to it in ōrbita and that nobody has removed your agents’ access to that company.
  • Your AI does not see a task. It has to be open (finished tasks are not on the list) and you need access to the tasks of that company in ōrbita (Editor level or higher).
  • It cannot put a file of the task on the post. Only images and videos within the limits above work, and the post cannot already be scheduled or published.
  • It says it has reached the usage limit. The connector has a limit of requests per minute and per hour to protect the service. Wait a little and try again.
  • It cannot edit a post. Posts that are already scheduled or published are locked for your AI just as they are for you in the app.
  • The authorization fails or expires. Start the connection again from your assistant and complete the authorization without switching browsers.

Privacy

ōrbita does not see your Claude or ChatGPT account, your conversations or your subscription. What it keeps is which AI you have authorized, with which permissions, and a log of which operation it ran, in which company and when; if it scheduled or replied at your request, also the executed instruction (text, network and date). Never the content of the conversation. What your AI reads from ōrbita goes to your assistant’s provider under the terms you have with it. More in the privacy policy.

Who it is for

The connector is included in the Satellite, Star and Galaxy plans, at no extra cost, for all the companies in the account. Your AI can do the same as you with each company’s plan: if a feature is not in its plan (for example, catalog, key dates or social proof on Satellite), or the company has been left without a plan, your AI cannot use it either. You bring the AI: the connector works with the subscription you already have with your assistant.

Published . Updated .