Connect your agent in a minute.

Build with Montage.

Montage gives your agent one MCP server. It searches posts, videos, articles and products across social media and the web, each with its original link and creator credit, understands videos by what is said and shown, and keeps watching what you choose so you can ask about it later, with a citation for every claim. A free account is all you need.

Connect

One server over Streamable HTTP. The first time your agent connects, your browser opens once to sign in with Google; that creates your free account and workspace, and the agent shares it with the app at app.montages.co. Headless agents send a workspace key instead.

claude mcp add --transport http montage https://app.montages.co/mcp

{
  "mcpServers": {
    "montage": { "type": "http", "url": "https://app.montages.co/mcp" }
  }
}

Other clients: add a remote server, choose Streamable HTTP and enter the URL. Try the tracking tools first with sample data and no account at https://app.montages.co/demo/mcp.

Search tools

Three read-only tools for searching and understanding what is out there. No modes or schemas to choose.

research

Search with a natural-language query. Montage picks the sources, ranks the results and answers within 5 seconds.

more_results

Get the next page of results, results from slower sources that arrived after the first answer, or the rest of a long transcript.

watch_video

Answer a question about one video with timestamped moments of what is said and shown. Each video is processed once, so later questions about it return in about a second.

Workspace tools

Workspace tools track the accounts you choose and answer questions about what they post and who responds. The app at app.montages.co and your agents share one workspace. Every write takes a request_key, so a retried call never runs twice. Every result has the same shape: {schema: "signals.v1", data, meta}.

signals_status

Free and read-only. What the workspace can do, its budgets, its watchlists, or one watchlist's accounts with coverage and freshness. Start here.

signals_discover

Suggests accounts for a brief, ranked, each with reasons and source links. Stores nothing.

signals_watch

Starts, changes or stops tracking. Tracking means Montage watches public posts; it never follows, subscribes or posts. Checks budget and plan limits before any effect and returns a job to poll with signals_status.

signals_query

Ask a question about what was collected. Answers cite evidence rows and report coverage; nothing is collected by asking.

signals_audience saves a reproducible list of accounts that meet explicit rules, signals_export turns it into CSV or JSON, and signals_next continues any result from meta.next_cursor.

Reading results

Search returns up to 15 results grouped by category under results; every category is present, even when empty. Each result keeps its title, original link, source and creator credit. Dates, excerpts and images appear when the source permits them.

{
  "results": {
    "videos": [
      { "title": "I Tried the $90 Lidl Creami (Is It Better Than My Ninja?)",
        "source": "youtube.com", "published_at": "2026-07-30" }
    ],
    "web_pages": [
      { "title": "People Are Finding Maggots Inside Ninja Creami Machines on TikTok",
        "source": "today.com", "published_at": "2026-08-19" }
    ],
    "images": [], "social_posts": [], "news": [], "products": []
  },
  "meta": { "status": "partial", "next_cursor": null }
}

Rate limits

Each workspace key and each signed-in workspace gets 60 tool calls per minute, counted across search and workspace tools. Connecting and listing tools don't count. When you hit the limit, wait for the time in the Retry-After header. Collection that tracking starts also spends the workspace's daily budget, which signals_status reports.

Errors

Useful partial results still return HTTP 200, with the gaps listed under meta.coverage.

HTTP API

Prefer plain HTTP? Every tool is also an HTTP endpoint, POST https://app.montages.co/api/tools/<tool name>, with the same arguments as over MCP and a workspace key.

curl -X POST https://app.montages.co/api/tools/research \
  -H 'Authorization: Bearer <WORKSPACE_KEY>' \
  -H 'Content-Type: application/json' \
  -d '{"query":"Ninja Creami reviews and complaints","limit":8}'

curl -X POST "https://app.montages.co/api/tools/signals_query" \
  -H 'Authorization: Bearer <WORKSPACE_KEY>' \
  -H 'Content-Type: application/json' \
  -d '{"watchlist_id":"<WATCHLIST_ID>","question":"What are people complaining about?"}'

Agent-ready files

Point your agent at these and it can set itself up:

• montages.co/llms.txtShort reference
• app.montages.co/llms.txtEvery tool, from the server itself