Instagram MCP server for Claude Code and AI agents. Read your own accounts, research any Business or Creator account on the platform, publish (reels, stories and carousels from a video file on your machine, or from a URL), and answer comments.
Three tiers of access behind one tool surface, and every answer tells you which tier it came from.
Built by Navid Moazzez.
| Section | ||
|---|---|---|
| 1 | What you can ask it | Real prompts, not features |
| 2 | Quick install | The package, no account needed |
| 3 | Create your Meta app | Every click. This is the part people give up on |
| 4 | Get your token | One command does the exchange |
| 5 | Connect your client | Claude Code, Desktop, Cursor, the rest |
| 6 | Check it worked | doctor, and the two things that actually fail |
| 7 | Tools | All 47, by what they reach |
| 8 | The three tiers | What each one can see |
| 9 | Multiple accounts | One server, several logins |
| 10 | Notes and gotchas | Quotas, windows, and silent failures |
| 11 | Troubleshooting | Symptom to cause |
| 12 | FAQ | Including what an MCP server is |
- Which of my posts from the last month is still gaining likes?
- How many followers did I gain this week, and which post caused it?
- Compare @a, @b and @c. Who has the best engagement rate, and what do they post?
- What is working on #aitools right now?
- Read every comment from my last ten posts and tell me what people keep asking.
- Draft a reply to each comment that deserves one, in my voice. Do not post them.
- Stage this carousel as a container so I can look at it before it goes live.
- Post ~/Movies/launch.mp4 as a reel with this caption.
- How many of today's 100 API posts have I used?
The first two are the point. Instagram will tell you your follower count right now. It will not tell you what it was on Monday. This server remembers what it reads, so it can answer the version of the question people actually have.
Note
Not on PyPI yet. Install it from a checkout until it is published. The command below is what it will become, and the rest of this README already works against a local install.
Python 3.11 or newer.
git clone https://github.com/thenavidm/instagram-mcp
uv tool install ./instagram-mcpNo uv yet? It is a Python package manager, one command to install:
curl -LsSf https://astral.sh/uv/install.sh | shCheck it:
instagram-mcp --versionOnce published:
uv tool install thenavidm-instagram-mcpThe package name and the command differ on purpose. The short name on PyPI
belongs to somebody else, so installing that would fetch code that is not this.
The installed command is instagram-mcp either way.
For the unofficial tier as well, read section 8 first, then add the extra:
uv tool install "./instagram-mcp[unofficial]"Tip
One app covers Facebook, Instagram and Threads.
Use cases are ticked in a list, and you can tick several. If you plan to use more than one of these, do it now rather than making three apps and managing three sets of credentials.
| Use case | For | Server |
|---|---|---|
| Manage everything on your Page | Facebook Pages | facebook-mcp |
| Manage messaging and content on Instagram | this one | |
| Access Threads API | Threads | threads-mcp |
Incompatible combinations grey out. If an option will not tick, it conflicts with something already selected.
Instagram's API does not hand out tokens directly. You create an app in Meta's developer dashboard, and the app issues the token. It is free and takes about ten minutes.
Before you start: your Instagram account must be a Business or Creator account, not a personal one. Instagram's API does not work on personal accounts at all, on any tier except the unofficial one. Switch in the Instagram app under Settings, then Account type and tools.
-
Go to developers.facebook.com/apps/creation and log in.
-
App details. Enter your app's name and a contact email address, then Next. The name is not shown to anyone but you at this stage.
-
Use cases. This screen is a list of what the app is allowed to do. Tick Manage messaging and content on Instagram, then Next.
That one name covers reading posts, insights, comments, DMs and publishing. There is no separate "Instagram API" option, which is what people look for and do not find.
The other names on this screen belong to different products:
Use case What it is for Manage messaging and content on Instagram this server, tick it Access Threads API Threads, see below Manage everything on your Page Facebook Pages You can tick more than one. Incompatible combinations are greyed out, so if an option will not tick, it conflicts with something already selected.
Important
Older guides tell you to pick "Other" and then choose an app type of "Business". Those options are not on this screen.
Meta reorganized app creation around use cases, so instead of picking a generic type and adding a product afterwards, you pick the thing you want to do and Meta adds the products for you. Almost every Instagram tutorial online still describes the old flow.
If you are looking for "Other", you are on the right screen. Tick Manage messaging and content on Instagram instead. There is no app type step after it.
Tip
You do not need a callback or redirect URL for this server.
A redirect URL is for a hosted login flow, where other people sign into your app and get sent back to your website. That is not what is happening here. You are generating a token for your own account, from your own dashboard, and the server runs on your machine.
The Facebook Login setup does not ask for one at all. The Instagram Login setup has a Business login settings section with OAuth redirect URIs, deauthorization and data deletion URLs. You can leave those alone unless you are building a website other people log into.
If a guide tells you to host a callback URL to use an MCP server, it is describing a different job.
- Business. Choose A verified business portfolio, An unverified business portfolio, or I don't want to connect a business portfolio yet. The last option is fine to start with.
- Requirements. Review whatever it lists, then Next.
- Overview. Check the details, then Go to dashboard.
- Click Dashboard in the left menu.
- Select the use case to customise.
- Click Add all required permissions.
You can also reach these under Permissions and features in the left menu, to add or remove individual ones.
Two setups, and they reach different things. This is the decision that matters most, so read the table in section 8 before choosing.
Instagram Login is the simpler one. In the left menu, API setup with Instagram login. It reaches only accounts you own.
| Permission | For |
|---|---|
instagram_business_basic |
reading the account |
instagram_business_content_publish |
publishing |
instagram_business_manage_comments |
comments |
instagram_business_manage_messages |
DMs |
Facebook Login reaches more. It is the only path that gets
discover_account and the hashtag tools, and the Page tokens it returns do not
expire. It needs your Instagram account linked to a Facebook Page.
| Permission | For |
|---|---|
instagram_basic |
reading the account |
instagram_content_publish |
publishing |
instagram_manage_comments |
comments |
instagram_manage_insights |
reach, saves, impressions |
instagram_manage_messages |
DMs |
pages_show_list |
finding the Page |
pages_read_engagement |
reading the linked Page |
business_management |
required by Meta for the messaging and content setups |
In the use case's Generate access tokens section, click Add account and approve the dialog. Meta returns a short-lived token.
For the Facebook Login path you can also use Graph API Explorer: select your app, add the permissions above, then generate.
Copy the token. It is short-lived, which is fine: section 4 trades it for a long-lived one.
Tip
Your app does not need review to work. App Review is only needed when people outside your app's Roles list will use it. For your own accounts, skip Go to app review entirely.
One command does the exchange:
instagram-mcp token \
--app-id YOUR_APP_ID \
--app-secret YOUR_APP_SECRET \
--short-token THE_SHORT_LIVED_TOKEN \
--path facebookIt swaps the short-lived token for a long-lived one, finds every Page you manage
with a linked Instagram account, and prints a ready-to-paste IG_ACCOUNTS_FILE.
Underneath, the Instagram Login exchange is this:
GET https://graph.instagram.com/access_token
?grant_type=ig_exchange_token
&client_secret=YOUR_APP_SECRET
&access_token=THE_SHORT_LIVED_TOKEN
That returns a token good for 60 days. The request carries your app secret, so it belongs in server-side code only. Never in a browser, never in anything shipped to a device, never committed.
Extend it before it expires:
instagram-mcp refreshFacebook Login Page tokens do not expire, which is the other reason to prefer that path.
You do not create a second app for those. The same app carries several use cases, with one app id, one secret and one Roles list, so a tester you add once can use all of them.
What is not shared is the token. Each product mints its own against its own host, so an Instagram token does not work on Threads.
If you want either, tick its use case back in step 1 and then follow that server's own setup, which is written up in its repo rather than duplicated here:
| You also want | Tick | Then |
|---|---|---|
| Threads | Access Threads API | a Threads server is in progress, not published yet |
| Facebook ad research | nothing, the Ad Library is public and needs no app | facebook-ad-library-mcp |
Adding a use case later is fine. It does not invalidate the token you already have.
Writes are on. There is no flag to enable publishing or replying. The actions
that cannot be undone from a chat window ask the model to pass confirm: true
first, and IG_READ_ONLY=1 removes every write tool if you want a reader.
claude mcp add --transport stdio instagram \
--env IG_ACCESS_TOKEN=your_token \
--env IG_USER_ID=your_ig_user_id \
-- instagram-mcpBy default this applies to the current project only. Add --scope user to use
it everywhere.
| Command | What it does |
|---|---|
claude mcp list |
List every configured server |
claude mcp get instagram |
Show this server's status |
/mcp |
Check the connection from inside a session |
| Platform | Config path |
|---|---|
| macOS | ~/Library/Application Support/Claude/claude_desktop_config.json |
| Windows | %APPDATA%\Claude\claude_desktop_config.json |
{
"mcpServers": {
"instagram": {
"command": "/Users/you/.local/bin/instagram-mcp",
"env": {
"IG_ACCESS_TOKEN": "your_token",
"IG_USER_ID": "your_ig_user_id"
}
}
}
}Tip
The path must be absolute. Claude Desktop does not inherit your shell PATH, so
a bare command name fails silently. Get it with which instagram-mcp on macOS
or where instagram-mcp on Windows, and escape the backslashes on Windows.
Then quit Claude Desktop completely and reopen it. Closing the window is not enough.
.cursor/mcp.json:
{
"mcpServers": {
"instagram": {
"command": "instagram-mcp",
"env": {
"IG_ACCESS_TOKEN": "your_token",
"IG_USER_ID": "your_ig_user_id"
}
}
}
}Run it over HTTP:
instagram-mcp --http --host 0.0.0.0 --port 8000Then add https://your-host/mcp as a custom connector.
Put it behind TLS and an authenticating proxy. This server holds tokens for your Instagram accounts and has no authentication of its own.
All of them take the same three things: the command instagram-mcp, any flags
as arguments, and the environment variables above. See
docs/clients.md for the exact file and key names.
instagram-mcp doctordoctor is the whole troubleshooting story. It tests every token, names which
insight metrics your account still answers, and tells you whether
business_discovery works on your app. Run it before anything else.
Two things fail more than everything else combined:
A personal account. The API returns an empty or confusing error rather than
saying so. doctor says so plainly.
A token generated without the scopes. Adding scopes in Graph API Explorer does not update a token you already copied. Generate a new one after adding them.
Every tool also runs as a command, through the same server and the same
confirmation as over MCP. instagram-mcp call <tool> prints the tool's JSON
result, and its exit code is what a script, cron or a scheduler branches on.
instagram-mcp tools lists them, marking with ! the ones that need
--confirm.
instagram-mcp tools # every tool, one line each
instagram-mcp call get-publishing-limit --account yourbrand
instagram-mcp call publish-reel --account yourbrand \
--file /Users/you/Movies/launch.mp4 --caption "It's live" --confirm --agent
instagram-mcp call create-container --account yourbrand \
--args '{"file": "/Users/you/Movies/launch.mp4", "caption": "Later"}'- A tool's arguments are
--name value, with dashes or underscores. A boolean alone means true; a list takes the flag once per item, or a JSON list.--argstakes them all as one JSON object, and flags win over it. --confirmis the only way to confirm from the shell.confirminside the arguments is refused, and--agent(compact JSON on one line) never adds it. A tool with nothing to confirm ignores--confirm, so a scheduler can pass it to every command.- A failure is one JSON object on stderr,
{"error": ..., "code": ...}.
| Exit | Means |
|---|---|
| 0 | Done |
| 1 | Unexpected error |
| 2 | Usage: an unknown tool or flag, a bad value, or a write refused (read-only, or no --confirm) |
| 3 | Not found: an account name that is not connected, or a Graph 404 |
| 4 | Auth: a token Meta refuses or a missing permission |
| 5 | Meta failed, or did not answer |
| 7 | Rate limited, wait and retry |
| 10 | Nothing configured: no account, or the unofficial tier off |
47 tools. Every result carries a source field naming the tier that answered,
so a model can never present a scraped guess as an official metric.
| Tool | What it does |
|---|---|
list_accounts |
Every connected account with live follower counts and which tier it reaches |
whoami |
Verify one token, return the live profile |
token_status |
When each token expires and whether it can be refreshed |
get_media |
Recent posts with permalinks and engagement |
list_all_media |
Every post, paging until the history is exhausted |
get_media_by_id |
One post, full fields |
list_tagged_media |
Posts by other accounts that tagged you |
list_stories |
Stories live right now |
get_media_insights |
Reach, saves, shares and interactions for one post |
get_account_insights |
Account-level reach and profile views over a period |
get_publishing_limit |
How many of the 100 daily API posts you have used |
growth_history |
Follower counts over time, from local readings |
post_movement |
Which posts are still gaining, and by how much |
growth_history and post_movement read the local store, not Instagram. They
are empty until the server has taken readings on more than one day, and they
cannot be back-filled. This is the reason to install it before you need it.
| Tool | Needs confirm |
|---|---|
create_container |
no, nothing goes public |
get_container_status |
read |
search_audio |
read |
publish_container |
yes |
post |
yes |
post_carousel |
yes |
publish_reel |
yes |
publish_story |
yes |
A reel can carry a song from Instagram's own licensed library: search_audio finds it (what is trending
now when no query is given, or any song or sound by name) and audio_id on publish_reel or create_container
attaches it, through Meta's Audio API, which answers on the Facebook Login path only. audio_volume and
video_volume (1 to 100) set the song against the clip's own sound. Meta takes a song on reels only: a carousel
or a photo refuses it.
create_container stages media without posting. Nothing appears in the app and
the container expires unused after 24 hours. It is the only draft-like state
Instagram has, so it is what to use when a person should approve something
before it goes live.
Instagram has no native scheduling. Anything that claims to schedule a post is
holding the job somewhere else and publishing it at the time. So there are two
ways to publish: now, with post, post_carousel, publish_reel or
publish_story, or staged, with create_container, then
get_container_status until it says FINISHED, then publish_container
within the 24 hours.
Every publishing tool takes file, a path on this machine, next to the URL
arguments, and post_carousel takes items, URLs and paths mixed, in order.
A video file goes up with Meta's resumable upload: a container opened with
upload_type=resumable, then the bytes sent to rupload.facebook.com, as
Meta's content publishing guide
describes. The file is streamed from disk, and a dropped connection is sent
again, up to three times. Meta documents no way to ask how far an Instagram
upload got, so it starts again from the first byte rather than guessing.
Two limits are Meta's, and both are checked before anything is sent:
- Facebook Login only. The guide lists
rupload.facebook.comas a host for Facebook Login, and says resumable upload is "Only for apps that have implemented Facebook Login for Business". An account on the Instagram Login path still publishes from a URL;instagram-mcp token --path facebookconnects it the other way. - Video only. Meta's resumable upload takes reels, video stories and video
carousel items. An image container takes
image_urland nothing else, so an image file is refused with that reason: host the JPEG and pass its URL.
| Tool | file takes |
Without a media_type |
|---|---|---|
post |
a video | posts a reel; media_type: STORIES for a story |
create_container |
a video | stages a reel; STORIES, or is_carousel_item for a carousel child |
publish_reel |
a video | a reel |
publish_story |
a video | a story |
post_carousel |
items: image and video URLs, video paths |
each item in order |
| Limit | Meta's number | Source |
|---|---|---|
| API-published posts per account | 100 in a 24-hour moving period, a carousel counting as one. The carousel section of the same page says 50; get_publishing_limit reads the number Meta applies to your account |
Content publishing |
| Containers | Expire after 24 hours unpublished; 400 created per account in a rolling 24 hours | IG User Media |
| Carousel | 2 to 20 images and videos, no reels | Content publishing |
| Reel | MOV or MP4, 3 seconds to 15 minutes, 300 MB; 9:16 recommended | IG User Media |
| Story video | 3 to 60 seconds, 100 MB | IG User Media |
| Image | JPEG only, 8 MB | IG User Media |
| Tool | Needs confirm |
|---|---|
get_comments |
read |
get_comment_replies |
read |
read_all_comments |
read |
list_conversations |
read |
get_conversation |
read |
reply_to_comment |
no, a reply can be deleted |
hide_comment |
no, hiding is reversible |
delete_comment |
yes, it is permanent |
send_dm |
yes, a message cannot be unsent |
private_reply_to_comment |
yes, and only one is allowed per comment |
read_all_comments pulls comments across your recent posts in one call. It is
what comment triage and sentiment questions actually need, and doing it through
get_comments in a loop burns the context window before it answers anything.
| Tool | What it does |
|---|---|
discover_account |
Public profile and recent posts for any Business or Creator account |
compare_accounts |
Up to 10 accounts side by side with median engagement |
search_hashtag |
Resolve a hashtag to its id, remembering the result |
hashtag_top_media |
What is performing on a hashtag right now |
hashtag_recent_media |
What is being posted on a hashtag right now |
These need the Facebook Login path. discover_account is an official, free,
documented Graph endpoint. No scraping, no risk, no browser.
Off unless you turn it on. See section 8.
| Tool | What it does |
|---|---|
unofficial_status |
Whether the tier is on, and what is deliberately not implemented |
unofficial_profile |
Full public profile for any account, including personal ones |
unofficial_posts |
Recent posts by any public account |
unofficial_stories |
Stories live on any public account |
unofficial_followers |
Followers, newest first, capped at 200 |
unofficial_following |
Accounts an account follows, capped at 200 |
unofficial_search_accounts |
Search accounts by name or keyword |
unofficial_post_comments |
Comments on any public post |
unofficial_inbox |
Your real inbox, including message requests |
unofficial_thread |
Messages inside one inbox thread |
unofficial_send_dm |
A cold DM. Needs confirm |
Three things are deliberately absent: follow_user, unfollow_user and
bulk_like. They are what every growth-hack tool ships and they are the three
actions that get accounts restricted fastest. unofficial_status reports them
as missing on purpose, so it is a decision on the record rather than an
oversight.
Instagram does not have one API. It has two, plus a private one, and they reach completely different things.
| Tier | What it reaches | Runs where | Risk |
|---|---|---|---|
| Instagram Login | only accounts you own | anywhere | none |
| Facebook Login | the above, plus public data on any Business or Creator account, plus hashtags | anywhere | none |
| Unofficial | any public account, the real inbox, account search | a machine you control | your account can be restricted |
Most people want the middle one and do not know it exists.
Warning
The unofficial tier drives Instagram's private API through instagrapi. That
is against Instagram's terms of service and it can get an account restricted
or banned. Point it at a secondary account, never your main one.
It is not installed by default, it is off unless you set IG_UNOFFICIAL=1 or
pass --unofficial, and it prints a warning on stderr when it starts.
It also paces itself: a random delay between calls and a local hourly ceiling that is lower than Instagram's own. Instagram restricts accounts for machine-speed access patterns more than for volume.
instagram-mcp login # once, saves a session file
IG_UNOFFICIAL=1 instagram-mcpPoint IG_ACCOUNTS_FILE at a JSON array instead of setting a single token.
Mixed tiers in one file are fine and expected.
[
{
"account_name": "acmem",
"user_id": "17841400000000000",
"access_token": "IGQV...",
"host": "graph.instagram.com"
},
{
"account_name": "acmeai",
"user_id": "17841400000000001",
"access_token": "EAAG...",
"host": "graph.facebook.com"
}
]Every tool takes an optional account argument. When you leave it out, the
IG_PREFERRED order decides, and an exact name match beats a prefix match.
IG_PREFERRED=acmem,acmeai- A video can come from your machine; an image cannot. A video
fileuploads through Meta's resumable upload on a Facebook Login account. An image has to be at a public URL, because Meta fetches it server-side and documents no upload for one. - 100 posts per 24 hours, per account, across every API client, not just this
one, by the publishing guide's rate limit section; its carousel section says
50.
get_publishing_limittells you which applies and where you are. - Containers expire after 24 hours. A staged post that is not published by
then has to be staged again.
get_container_statussaysEXPIRED. - 30 unique hashtags per rolling 7 days. Resolved ids are cached locally so repeat lookups do not spend a slot, but the thirty-first new hashtag fails.
business_discoverysees Business and Creator accounts only. Never personal, never private. The error does not say which of those it hit.- DMs need a 24-hour window. Instagram allows a DM only inside a window the other person opened by messaging you, or as one private reply to a comment. There is no official way to message somebody cold.
- Reading DMs needs Advanced Access from Meta App Review on most apps.
list_conversationsreturns empty rather than erroring until you have it. - Insight metrics change. Meta deprecates and renames them regularly.
doctornames the ones your account still answers, which is faster than reading the changelog. - Errors come back as a structured result, not an exception, carrying the
Graph error's
code,subcodeandfbtrace_id. That is Instagram's error, not a bug in this server. - Comments and DMs are written by strangers. Every text field from another person comes back wrapped in an untrusted-content fence telling the model to report it, not obey it. This is the most injectable surface an agent gets handed, and "summarise my comments" is one of the first things anyone asks.
- Every write is logged. One JSON line per attempted write, allowed or
refused, to
IG_AUDIT_LOGor the data directory. The model has no tool to read or edit that file. - Tokens stay out of the logs. A Graph URL carries the token in its query string, and the HTTP library logs each URL. Those lines are off, and a filter removes any token from them if someone turns them back on.
Run instagram-mcp doctor first. It diagnoses most of this.
| Symptom | Cause |
|---|---|
| Every tool returns nothing | Personal account. The API needs Business or Creator |
(#100) Tried accessing nonexisting field |
Token generated before you added the scopes. Generate a new one |
discover_account fails |
You are on the Instagram Login path. It needs Facebook Login |
| Hashtag tools stop working | 30 unique hashtags in a rolling 7 days |
| Server missing in Claude Desktop | Relative command path. Use the absolute one, and fully quit the app |
growth_history is empty |
It needs readings from more than one day. It cannot back-fill |
| Write tools are absent | IG_READ_ONLY is set |
| A write asks for confirmation | Working as intended. Call it again with confirm: true |
A file refuses with "Facebook Login path" |
The account uses Instagram Login, which Meta gives no upload host. Use a URL, or connect it with token --path facebook |
A file refuses as an image |
Meta uploads video files only. Host the image and pass image_url |
upload 401 NotAuthorizedError |
The token lacks instagram_content_publish, or the app is not on Facebook Login for Business |
| Publishing says a container is still processing | A large video. get_container_status, then publish_container once FINISHED, within 24 hours |
| Unofficial tools refuse | Tier is off. Set IG_UNOFFICIAL=1 and run instagram-mcp login once |
doctor or list_accounts says Tried accessing nonexisting field (account_type) |
0.3.0 or earlier on a Facebook Login account. account_type exists only on Instagram Login; 0.4.0 asks each path only for fields it has |
What is an MCP server?
Model Context Protocol is a standard way to give an AI assistant access to a tool or a data source. An MCP server exposes a set of functions, and a client like Claude Code or Claude Desktop can call them during a conversation. This one exposes Instagram.
You install it once, put your credentials in the client's config, and then ask in plain language. You do not call the tools yourself.
Do I need a Business account?
You need one for everything except the unofficial tier. Instagram's API does not work on personal accounts. Switching is free and reversible, in the Instagram app under Settings, then Account type and tools.
Why do I have to create a Meta app? That seems like a lot.
Instagram does not issue API tokens to people, only to apps. Every tool that touches Instagram programmatically works this way, including the ones that hide it behind a signup form. Doing it yourself means the token is yours, it is not held by a third party, and nobody can revoke your access by shutting down.
It is about ten minutes, once.
Can it schedule posts?
It cannot, and neither can anything else. Instagram has no scheduling API. Products that offer scheduling hold the post on their own servers and publish it at the time, which means handing them your content and your credentials.
create_container is the honest version: stage it now, publish it when you say,
within the 24 hours Meta keeps a container.
Can it post a video from my computer?
Yes, on a Facebook Login account. Pass file with the path and it goes up
through Meta's resumable upload, as a reel, a story or a carousel item. Not an
image: Meta takes images only from a public URL. Instagram Login accounts
publish from a URL too.
Can it read my DMs?
It reads only the threads where the other person messaged you, only inside a 24-hour window, and only once Meta grants your app Advanced Access. The unofficial tier reads the real inbox with no window, at the risk described in section 8.
Will this get my account banned?
The two official tiers are Meta's own documented APIs. There is no more risk in using them than in using the Instagram app.
The unofficial tier is a real risk. It drives the private API, which is against Instagram's terms. Use a secondary account.
Can it post without asking me?
It can post when you ask it to. Publishing, deleting a comment and sending a DM
all require the model to pass confirm: true, which it does after reading a
description explaining what cannot be undone. That is a speed bump against a
careless call, not a lock.
If you want a server that cannot write at all, set IG_READ_ONLY=1. The write
tools are then not registered, so the model cannot see or call them.
What data does it store, and where?
It stores follower counts and post engagement readings in a SQLite file in your data directory, so it can answer "what changed since Monday". An audit log of every attempted write. A session file if you use the unofficial tier.
All of it is local. Nothing is sent anywhere except to Instagram. Delete the data directory and it is gone.
Why is this Python when the other servers are TypeScript?
The unofficial tier depends on instagrapi, which is Python only. Rewriting the
official tiers in TypeScript would split the project in two for no gain.
Why is the package name not just "instagram-mcp"?
That name on PyPI already belongs to somebody else. Installing it would fetch
code that is not this. The package is thenavidm-instagram-mcp and the command
it installs is instagram-mcp.
Does it cost anything?
It costs nothing. The server is MIT licensed and Meta's Graph API is free at these volumes. You are paying for your own AI client, not for this.
Can I use it with several accounts?
You can connect as many as you like, as section 9 covers. One server, one config
file, an account argument on every tool, and a preference order for when you
leave it out.
Run into a problem or have a question? Open an issue and I will help.
Navid Moazzez is a leading AI business strategist and the host of the AI Creator Summit, watched by 100,000+ creators. He helps creators and founders master AI and build their own AI Operating System (AI OS) to automate their business and life. He creates useful free tools, MCP servers and CLIs that creators and founders can use in their own workflows.
Links
- Personal website: navid.me
- Navid Media: navid.media
- YouTube: @thenavidm and @thenavidai
- X: @thenavidm
- Instagram: @thenavidm
- LinkedIn: thenavidm
If this is useful, star the repo and come say hi on X.
| Library | License | What it does |
|---|---|---|
| MCP Python SDK | MIT | The MCP server, stdio and streamable HTTP |
| httpx2 | BSD-3 | The HTTP client, already an SDK dependency |
| instagrapi | MIT | The unofficial tier, optional extra |
MIT. Free to use, modify, and share.
Not affiliated with, endorsed by, or sponsored by Meta Platforms, Inc. Instagram and Facebook are trademarks of Meta Platforms, Inc. This project uses Meta's public Graph API, and its optional unofficial tier is not sanctioned by Meta.
Β© 2026 Navid Media. Made with β€οΈ by Navid Moazzez.
