Instagram MCP server: let Claude set up your comment-to-DM automations
Connect Claude, ChatGPT, Codex, Cursor or any MCP client to CommentKeyword with one URL. Your agent can then create, re-target and pause Instagram keyword automations and check their results. When Claude edits a Reel that ends with "Comment GUIDE", it can arm that DM for your next post before you publish.
You:Edit this into a Reel. CTA: comment GUIDE to get the free checklist.
Claude:The Reel is edited. I added a "Comment GUIDE π" overlay inside the safe zone and put the CTA in the first line of the caption. Ready to arm the DM:
- Keyword: GUIDE (whole word)
- Posts: your next post or reel
- DM: "Here's the checklist, {{namefirst}} π https://β¦"
- Public reply: "Sent! Check your DMs π©"
Create it?
You:Yes
create_keyword Β· target: next_post
Claude:Done. GUIDE is waiting for your next post. Publish the Reel, then comment GUIDE from another account to test it.
What your AI agent can do with it
The MCP server turns the CommentKeyword API into tools your agent can call. You describe the campaign; the agent sets it up and shows you the result.
Set up automations from the chat
Ask for a keyword, the DM, variants and a public reply in plain language. The agent checks for a keyword that already exists, then creates the automation.
Arm the DM before you publish
With target: next_post you don't need a post ID. The automation attaches itself to the next post or reel the account publishes.
Hand the conversation to an AI agent
On paid plans, the agent can attach one of your Live AI agents with aiAgentId, so the conversation continues when the person replies to the DM.
When Claude edits your Reel, the DM is ready before you post
The comment-to-dm-cta Agent Skill switches on when Claude scripts, edits or captions a short video whose CTA asks viewers to comment a word. It follows the same steps every time:
- 1Finds the CTA in the script, transcript, on-screen text and caption. If the video promises a resource but never asks for a comment, it suggests a CTA.
- 2Picks a keyword that won't misfire: one distinctive word, not something people write anyway, such as YES or WOW.
- 3Makes the CTA consistent: said in the voiceover, shown as an overlay inside the Reels safe zone, and written in the first line of the caption.
- 4Writes the DM with the promised link, plus optional wording variants and a public reply.
- 5Shows you a summary and, only after you confirm, creates the automation for your next post or reel.
Using claude-code-video-toolkit? The /comment-dm add-on reads the CTA from your project and arms the DM before you publish.
Connect your AI app
Add the server by its URL, sign in with CommentKeyword and approve what the app may do. No API token, no Node.js and no config file to edit by hand.
Server URL
https://commentkeyword.com/connect/mcpClaude: claude.ai, desktop app, Cowork and mobile
Add it once to your Claude account and it works everywhere you use Claude, Cowork included.
- Open Claude's connectors page (Customize β Connectors), click + and choose to add a custom connector.
- Name it CommentKeyword and paste the server URL.
- Keep the options Claude marks as Detected: sign in now, with Claude's published identity. Don't add request headers.
- Click Add, sign in with CommentKeyword and approve what Claude may do.
- In a chat or a Cowork task, click + β Connectors and check that CommentKeyword is on.
On Team and Enterprise plans, an owner adds the connector under Organization settings β Connectors, then each member clicks Connect. Free plans can add one custom connector. If Claude Desktop shows "Account mismatch", the browser is signed in to a different Claude account: sign in to claude.ai there with the account you use in the app, or add the connector from claude.ai with that account.
Claude Code
Run this command in your terminal. Then open Claude Code, type /mcp, choose commentkeyword and sign in. With --scope user it works in every project.
Terminal
claude mcp add --transport http --scope user commentkeyword https://commentkeyword.com/connect/mcpCodex: CLI, IDE extension and app
Run this command in your terminal. Codex opens your browser so you can sign in with CommentKeyword; if it doesn't, run codex mcp login commentkeyword. The Codex IDE extension and app use the same ~/.codex/config.toml, so one setup covers all of them.
Terminal
codex mcp add commentkeyword --url https://commentkeyword.com/connect/mcpOr add it to ~/.codex/config.toml
[mcp_servers.commentkeyword]
url = "https://commentkeyword.com/connect/mcp"ChatGPT
- Open chatgpt.com/plugins, click + and choose Add custom MCP server.
- Name it CommentKeyword, paste the server URL under Connection and keep OAuth as the authentication.
- Confirm, create it and sign in with CommentKeyword.
Whether you can add custom MCP servers depends on your ChatGPT plan and, in a workspace, on its admin settings.
Cursor and other apps
In Cursor, add this to ~/.cursor/mcp.json (or to .cursor/mcp.json for one project) and sign in when Cursor asks. Any other app that connects to remote MCP servers with OAuth works with the same URL.
mcp.json
{
"mcpServers": {
"commentkeyword": {
"url": "https://commentkeyword.com/connect/mcp"
}
}
}Or run it on your computer with an API token
For apps that can't connect by URL, or if you prefer a token. Create one in CommentKeyword under Settings β API & MCP with keywords:read, keywords:write and agents:read, and add keywords:delete only if the agent should be able to delete automations. The local server has the keyword and AI agent tools; the stats, usage and conversation tools are on the remote server for now.
Claude Code
The plugin installs the skill and the MCP server. Claude Code asks for your API token when you enable it and keeps it in your system keychain. For the MCP server only, use claude mcp add and replace YOUR_TOKEN with your token.
Plugin: MCP server + Skill
/plugin marketplace add commentkeyword/instagram-mcp-server
/plugin install commentkeyword@commentkeywordMCP server only
claude mcp add --transport stdio --env COMMENTKEYWORD_API_TOKEN=YOUR_TOKEN commentkeyword -- npx -y github:commentkeyword/instagram-mcp-server#v0.2.0Claude Desktop
Open Settings β Developer β Edit Config, add the server to claude_desktop_config.json, replace YOUR_TOKEN with your token and restart Claude Desktop.
claude_desktop_config.json
{
"mcpServers": {
"commentkeyword": {
"command": "npx",
"args": ["-y", "github:commentkeyword/instagram-mcp-server#v0.2.0"],
"env": { "COMMENTKEYWORD_API_TOKEN": "YOUR_TOKEN" }
}
}
}Cursor
Add the server to ~/.cursor/mcp.json to use it in every project, or to .cursor/mcp.json for a single project. Replace YOUR_TOKEN with your token.
mcp.json
{
"mcpServers": {
"commentkeyword": {
"command": "npx",
"args": ["-y", "github:commentkeyword/instagram-mcp-server#v0.2.0"],
"env": { "COMMENTKEYWORD_API_TOKEN": "YOUR_TOKEN" }
}
}
}Windsurf
In the Cascade panel, open the MCP settings and open the MCP config file. Add the same mcpServers entry and replace YOUR_TOKEN with your token.
mcp_config.json
{
"mcpServers": {
"commentkeyword": {
"command": "npx",
"args": ["-y", "github:commentkeyword/instagram-mcp-server#v0.2.0"],
"env": { "COMMENTKEYWORD_API_TOKEN": "YOUR_TOKEN" }
}
}
}Codex
Add the server with the Codex CLI, or add the table to ~/.codex/config.toml (or to .codex/config.toml in a trusted project). Replace YOUR_TOKEN with your token.
Codex CLI
codex mcp add commentkeyword --env COMMENTKEYWORD_API_TOKEN=YOUR_TOKEN -- npx -y github:commentkeyword/instagram-mcp-server#v0.2.0~/.codex/config.toml
[mcp_servers.commentkeyword]
command = "npx"
args = ["-y", "github:commentkeyword/instagram-mcp-server#v0.2.0"]
[mcp_servers.commentkeyword.env]
COMMENTKEYWORD_API_TOKEN = "YOUR_TOKEN"The local server runs on your computer through npx, so it needs Node.js 18 or later.
Only want the skill? Download the skills/comment-to-dm-cta folder from the repository, zip it and upload it in Claude under Customize β Skills. Without the MCP server it still plans the CTA and the DM.
MCP tools
Every tool uses the permissions you approved when you connected (or your token's, on the local server), so the agent can only do what you allowed.
| Tool | What it does | Permission |
|---|---|---|
| create_keyword | Creates an automation. target is next_post, all_posts or specific_posts. Also takes DM variants, optional public replies, link tracking and an optional AI agent. | keywords:write |
| update_keyword | Changes any field. Re-arm it for the next upload with target: next_post, or pause it with isActive: false. | keywords:write |
| list_keywords | Lists automations, with an optional search, so the agent can spot a keyword that already exists. | keywords:read |
| get_keyword | Shows one automation, including whether the next post was caught (nextPostStatus). | keywords:read |
| delete_keyword | Deletes an automation. If an AI agent is attached, it first returns an impact summary for you to confirm. | keywords:delete |
| list_agents | Lists the AI agents in the workspace, so a Live agent can be attached with aiAgentId. | agents:read |
| get_keyword_stats | Results of one automation for a date range: DMs sent, replies, AI conversations, goals reached and link clicks. Counts only, no commenter data. | keywords:read |
| get_stats_overview | The same counts for all your automations together, plus the 5 keywords that sent the most DMs. | keywords:read |
| get_usage | Shows your plan, the DMs used this month, your active keywords and anything that stops DMs from going out, such as a used-up Free quota or Instagram needing to be reconnected. | keywords:read, agents:read |
| list_conversations | Lists the DM conversations your AI agents handled, with the data each one collected, such as a lead's email. No transcripts. | conversations:read |
| get_conversation | Shows one conversation: the contact, the collected data and the messages. | conversations:read |
It asks before it messages anyone
- Every automation messages real people. The skill and the tool descriptions tell the agent to confirm the keyword, the DM and the target with you before creating or activating anything.
- Each connection and each token works for one workspace, and you can disconnect or revoke it at any time in Settings β API & MCP.
- Tokens have scoped permissions. A token with only
keywords:readcan look at your automations but not change them. - Deleting a keyword that has an AI agent attached needs a second, explicit confirmation.
- DMs go out through Meta's official Instagram API, with the same rules as automations you create in the dashboard.
Included on every plan
The API and the MCP server are included on every plan, Free too. Your plan's normal limits still apply: Free includes 300 automated DMs a month and 3 active keywords. AI agents that continue the conversation after the first DM are available on paid plans.
Frequently asked questions
What is the CommentKeyword MCP server?
A Model Context Protocol server that turns CommentKeyword into tools an AI agent can call: create, update, list, read and delete keyword automations, check their results and your plan usage, list your AI agents and read their conversations. Apps connect to the hosted server at https://commentkeyword.com/connect/mcp by URL, and an open-source (MIT) version runs on your computer.
Does it work with ChatGPT, Cursor or other AI tools?
Yes. Claude (Cowork included), Claude Code, ChatGPT, Codex, Cursor and other apps that connect to MCP servers by URL use the remote server at https://commentkeyword.com/connect/mcp. Apps that can only start local servers can run the open-source server with an API token. The Agent Skill is a standard SKILL.md folder, so other agents that read Agent Skills can use it too.
Does it use OAuth?
The remote server does: you sign in with CommentKeyword and approve what the app may do, and you can disconnect it at any time in Settings β API & MCP. The local server uses a personal access token that you create in Settings β API and pass as COMMENTKEYWORD_API_TOKEN. The Claude Code plugin keeps the token in your system keychain. You can revoke a token at any time.
What do I need on the Instagram side?
An Instagram professional account (Business or Creator) connected to CommentKeyword. CommentKeyword sends the DMs through Meta's official Instagram API. The agent never logs in to Instagram.
Can the agent start sending DMs without asking me?
The skill and the tool descriptions tell the agent to show you the keyword, the DM and the target, and to wait for your OK before creating or activating anything. If you only want it to look, give it a token with read permissions only. After you confirm, the automation works like one created in the dashboard: only people who comment the keyword get the DM.
Do I need the post ID?
No. With the next_post target, the automation attaches itself to the next post or reel the account publishes. That is the usual case while a video is still being edited. You can also target all posts, or specific posts by their Instagram media ID.
Does it cost extra?
No. The API and the MCP server are included on every plan, including Free. Automated DMs count toward your plan's monthly allowance as usual.
Is there an official Instagram MCP?
No. As of October 2026, Meta offers an official MCP server for ads, not for organic Instagram posts, comments or DMs, so every Instagram MCP server is third-party. This one works through CommentKeyword, which uses Meta's official Instagram API and never your password. See how Instagram MCP servers compare.
Is there a ManyChat MCP server?
Not an official one, as of October 2026. ManyChat's API can't create comment automations, so third-party ManyChat MCP servers can't either. Does ManyChat have an MCP server?
Let your agent arm the DM, then publish
Create a free CommentKeyword account, connect your Instagram professional account, then connect your AI app with the server URL.
New to this? See what Claude can and can't do with InstagramPrefer plain HTTP? See the Instagram automation API