InsightSocial MCP server

insightsocial

The official CLI and MCP server for the InsightSocial API: live public data from 22 platforms through one key: TikTok, LinkedIn, YouTube, X, Reddit, Pinterest, Bluesky, Truth Social, Snapchat, Telegram, Twitch, Kick, Rumble, Kwai, Douyin, Xiaohongshu (RedNote), Quora, TikTok Shop, Weibo, Zhihu, Substack and Nextdoor.

• 230+ endpoints: profiles, public emails, posts, videos, comments, followers, likers, hashtags, search, company pages, jobs, ads, transcripts, shop products and local business listings.
• One schema across platforms: a TikTok video and a YouTube video come back with the same fields (post.url, post.kind, post.engagement.likes, post.language), and unavailable lists any field a response could not fill. The CLI and MCP server ask for schema 2 on every call.
• Priced per endpoint, charged per call. Empty results, failed calls and idempotent replays are free.
• Built for agents. Every result is saved in full, and you re-slice it locally with jq for free, so a 50 KB page never ends up in your context.

Quick start

npx -y insightsocial init              # sign in through your browser, install the agent skill, show MCP setup
npx -y insightsocial search tiktok followers
npx -y insightsocial run /v1/tiktok/profile -p handle=natgeo
npx -y insightsocial view --last --jq '.data.author | {username, followers}'

init and login sign you in through your browser: they open the sign-in page and print a short code, you type the code there and click Allow, and a key named after this machine is saved. The CLI then prints which account it signed into. To use a key you already have, pass --api-key. New accounts include free calls.

Use it from an AI agent

Claude Code plugin

/plugin marketplace add insightsocial/skills
/plugin install insightsocial@insightsocial

This installs the skill and the MCP server together, from insightsocial/skills. Run npx -y insightsocial login once so the server can find your key.

MCP server (any client)

claude mcp add insightsocial --scope user -- npx -y insightsocial mcp

{
  "mcpServers": {
    "insightsocial": { "command": "npx", "args": ["-y", "insightsocial", "mcp"] }
  }
}

The server reads the key saved by insightsocial login, so the config holds no secret. You can also pass INSIGHTSOCIAL_API_KEY in its env. Without a key, search and describe still work.

Nothing to install: the same tools are hosted at https://api.insightsocial.app/mcp. Claude, ChatGPT and other clients that support it sign in with OAuth; others send the key as Authorization: Bearer. See MCP docs.

Tool	What it does
search_endpoints	Find endpoints by keywords and platform. Free.
describe_endpoint	Inputs, allowed values, an example and the price for one endpoint. Free.
call_endpoint	Fetch data. Saves the full response and returns a result_id with a trimmed view (10 items by default) and a ready next_call for the next page.
read_result	Re-slice a saved result with jq, fields, max_items or an outline with sizes. Free.
get_credits	Balance, plan and usage. Free.

insightsocial init --yes writes the MCP entry into Claude Code, Codex and Cursor for you.

Commands

Command
init [--yes] [--no-skills]	Save a key, install the skill for detected agents, print or write the MCP config.
login [--api-key k] [--no-browser] / logout	Sign in through your browser (or save a key you pass) / remove the key.
search <words> [--platform p]	Find endpoints. Free.
list [--platform p]	Every endpoint with its price. Free.
describe <path>	Inputs, example and price. Free.
run <path> -p name=value …	Call an endpoint. Saves to ./.insightsocial/. Options: --input <json>, -i file, --idempotency-key, --fresh, --max-credits N, -o file, --json, and the shaping flags below.
view [file] [--last [endpoint]]	Re-shape a saved result. No network, no charge.
credits	Balance and usage. Free.
mcp	Start the MCP server over stdio.

Shaping flags (run and view) trim only what is printed; the saved file is always complete:

• --jq <expr>: real jq (bundled, nothing to install) over the saved response. Paths start at .data.
• --fields a,b.c: keep these keys on each item. Dotted paths descend.
• --max-items N: show at most N items.
• --summary: the structure with byte sizes, so you can see what is large before pulling it.

Paging: when there is another page, run prints the exact next command. Each page is a separate charged call.

Pricing a call first: -p dry_run=1 (or true) on any endpoint returns the quote and charges nothing. When a response could not fill a field the platform normally has, run prints it under missing; that null means unknown, not zero.

Configuration

Key	--api-key, then INSIGHTSOCIAL_API_KEY, then ~/.insightsocial/config.json (mode 600)
INSIGHTSOCIAL_BASE_URL	Defaults to https://api.insightsocial.app
INSIGHTSOCIAL_HOME	Defaults to ~/.insightsocial (config, catalogue cache, MCP results)

Develop

npm install
npm run check      # build + unit tests, no network and no key needed

License

MIT