AI agents (MCP)

Run InstaVision from your AI agent

InstaVision's MCP server lets your AI agent find Instagram creators and leads for you: plan a search, estimate its cost, launch it with your API key and read the results. Runs it starts appear in your Runs list and spend your credits.

No key yet? The server answers list_playbooks and estimate_credits without one, so a remote client can plan and price a search before you create a key.

What your agent can do

Plan

No API key needed.

  • list_playbooks

    Lists the search playbooks: what each one is for, the fields it takes and its credit range.

  • estimate_credits

    Estimates what a run would cost, without launching anything or spending a credit.

Launch

Spends your credits.

  • launch_discovery

    Starts a run with a credit cap, the most the run may spend. Its MCP annotations mark it as destructive, so your client can ask you before each launch.

Follow and collect

Your runs, as they finish.

  • get_run_status

    Reports a run's progress (queued, running, succeeded, failed or aborted) with its counts and the credits charged.

  • get_run_results

    Returns the accounts a run found, page by page, filtered by email, follower range, category or relevance.

  • export_run_pdf

    Turns a run's results into a PDF and returns a download link that expires after 1 hour (or the file itself, when storage is unavailable).

Skip what you've seen

By default, runs skip accounts already delivered to you; a list you ask to enrich is scanned in full.

  • get_seen_accounts

    Lists the accounts already delivered to you or on your blocklist.

  • add_seen_accounts

    Adds handles to your blocklist so later runs skip them.

  • reset_seen_accounts

    Clears your blocklist, or your whole history of delivered accounts. It is marked as destructive and needs confirm=true.

Connect your agent

Clients that can send a header connect straight to the endpoint. Claude Desktop goes through a small local bridge, npx -y instavision-mcp.

Endpoint
https://instavision.co/api/mcp/mcp (Streamable HTTP)
Header
Authorization: Bearer <your API key>
Or
X-API-Key: <your API key> if your client can't set the Authorization header

Paste into a terminal (bash or zsh) and press Enter. It then waits for your key: paste it and press Enter. The key isn't shown or saved to your shell history; Claude Code stores it in ~/.claude.json and makes the server available in every project.

Terminal
read -rs INSTAVISION_API_KEY && claude mcp add --scope user --transport http instavision https://instavision.co/api/mcp/mcp --header "Authorization: Bearer $INSTAVISION_API_KEY"
ClientHow it connects
Claude CodeRemote, with your key in a header
CursorRemote, with your key in a header
CodexRemote, with your key in a header
VS CodeRemote, with your key in a header
Claude DesktopThrough the instavision-mcp npx bridge (needs Node.js)
claude.ai and ChatGPTNot yet: they sign in to MCP servers with OAuth, which InstaVision doesn't offer yet

Things to ask

  • “Find forex educators on Instagram who mention mentorship in their bio. Tell me what it will cost before you launch anything.”
  • “Find yoga teachers and studios in Lisbon, Portugal, and give me the ones with a public email.”
  • “Find more accounts like these: <paste a few handles>. Skip anyone I've already seen, then export the results as a PDF.”

Cost and safety

1 credit = 1 profile scanned

Runs your agent starts spend the credits of the account the key was created in: your own, or a team's.

You agree the cap

estimate_credits prices a run without spending anything, and launch_discovery's own description tells your agent that it spends real money and to agree the credit cap with you before launching. The cap is the most the run may spend, and a launch without one is refused.

Hard limits

An agent's launch can hold at most 500 credits. An agent can't launch while 5 runs of the same account are queued or running, website runs included. Agent launches may commit at most 2,000 credits per account in any 24 hours, counting their caps whatever they end up spending; launches from the website don't count toward it.

Short-lived links

PDF download links expire after 1 hour.

Revoke at any time

Revoke a key under Settings → API & MCP and the next request with it is refused. Runs it already started keep running.

Public data, untrusted text

InstaVision reads public Instagram profiles, never private accounts. Bios, names and captions are written by other people, and the tool that returns them tells your agent to treat them as data, never as instructions.

Questions

Do I need an API key?

To launch runs and read results, yes: create one under Settings → API & MCP (it is shown once). The server answers list_playbooks and estimate_credits without a key, so a remote client can plan and price a search before you have one.

What does a run cost?

1 credit = 1 profile scanned. Every launch carries a credit cap, the most the run may spend, and a launch from an agent can set it to at most 500 credits.

Can I connect from claude.ai or ChatGPT?

Not yet. claude.ai and ChatGPT sign in to remote MCP servers with OAuth, which InstaVision doesn't offer yet. Claude Code, Cursor, Codex and VS Code connect with your API key in a header, and Claude Desktop through the npx bridge.

How do I cut an agent off?

Revoke its key under Settings → API & MCP. The next request with that key is refused; runs it already started keep running.

Will my agent get accounts I've already seen?

Not by default: runs skip accounts already delivered to you, and always skip accounts on your blocklist (a list you ask to enrich is scanned in full). To include accounts you've already received, your agent can launch with excludeSeen set to false; reset_seen_accounts clears the blocklist.

InstaVision is an independent product, not affiliated with or endorsed by Instagram or Meta.