Connect Claude to Instagram. Then give it a real job.

Claude is the interface. Composio supplies the connector and Instagram tools. Your Instagram account supplies the data and permissions. A skill can describe a process; it does not grant access to an account.

• Use an Instagram Business or Creator account that you control.
• Check that your Claude plan and organization allow custom connectors. An organization owner may need to enable access.
• Review the current Composio access and pricing requirements before connecting. This guide does not promise free access.
• Start with reading posts. Account insights and publishing depend on the actual permissions and tools available.

Screenshots are from Composio’s documentation; they do not show a live Evolve Demand connection.

1. In Claude, open Customize → Connectors → Add custom connector.
2. Name it Composio. Enter https://connect.composio.dev/mcp as the server URL.
3. Follow the current authentication prompts and review the access requested.
4. In a chat with Composio enabled, ask: “Connect my Instagram account.” Complete the generated Instagram authorization for the intended account.

Checkpoint: a saved connector is only configuration. The next step verifies which account it can actually read.

Run a small, read-only job first. Check the returned handle and open one permalink yourself.

Use my Instagram connection to show the authenticated username and account ID. Retrieve my last 10 owned posts with media ID, permalink, timestamp, media_type and media_product_type. Do not publish, reply or modify anything. Flag missing fields. Let me verify one returned link before continuing.

Expected output: your account identity and up to 10 posts with links. An empty feed on a new account is not proof of a broken connection. An unrelated handle is a stop-and-correct result.

Compare Reels with Reels and carousels with carousels. Keep the raw counts next to any rate. A high save rate from a very small audience deserves context.

Use a fixed 30-day date range and state the dates and timezone. Follow pagination until the cohort is complete or clearly mark it partial. Group Reels separately using media_product_type, not only media_type=VIDEO. Retrieve supported saved and reach metrics for each item. Show IDs, permalinks, dates, raw saved, raw reach and saved/reach. Keep absent metrics absent; do not divide by zero. Compare within format, flag low-reach observations, and don't infer causation. Identify three caption themes worth a follow-up and write one content brief per theme. Don't claim to have watched videos you only have captions for.

Eligibility matters: the provider’s media-insights description inspected for this training lists at least 1,000 followers and media within two years. Verify the current connector’s requirements on your account; this is not a universal claim about Meta. If metrics are unavailable, use the post and comment workflows instead.

Expected output: a dated, format-separated evidence table and three proposed briefs. Missing data stays missing. Caption analysis cannot establish what happened throughout a video.

Choose owned posts in scope. Group what people actually ask, preserve the source IDs, and turn repeated questions into specific lessons.

For these owned media IDs, fetch comments through all pages in scope. Group recurring questions by theme. Return theme, count, representative comment IDs and a proposed educational post. Distinguish direct quotes from your interpretation. Draft three briefs for the most repeated themes. Do not send replies or DMs.

Business application: a salon’s recurring booking questions can inform a NailGrow content brief. A Shopify brand’s product-use questions can inform a ShopGrow demonstration. These are proposed applications, not reported client outcomes.

Expected output: question themes, counts and source IDs. Reading comments does not itself send replies or DMs.

The connector’s documented carousel route separates container creation from publication. Use the exact approved account, slide order and caption. First, prepare the review:

Use this approved account ID, caption and ordered set of image assets. Validate them against the current carousel-tool requirements. Show the target handle, slide order and caption for review. Do not create or publish anything yet.

After the account owner approves the exact materials, use this separate instruction:

Prepare the approved carousel with those exact assets and caption. Check the processing status. Publish only the approved container. Return the created media ID, fetch its permalink and verify the published post. If any tool fails, stop and report that failure; don't claim the post is live.

For the direct-URL path, the inspected tool requires public direct HTTPS JPEG URLs and documents 2–10 carousel items. A Drive share page is not an image URL. File inputs may also be available; inspect the live schema for your client.

Expected output: published media ID and a verified permalink. A container ID is not a published post. A prompt requesting approval is an operating instruction, not a guaranteed technical access control.