Research MCP Server — 10 Tools for AI Agents

Research MCP Server — 10 live research tools for AI agents through one endpoint

Connect Claude, Cursor, Claude Code, or ChatGPT to arXiv, Reddit, GitHub, Hacker News, OpenAlex, Wikipedia, CoinGecko, GDELT global news, Telegram, and Stack Exchange — with one MCP URL, no per-source API keys, and no separate integrations to wire up.

What it does

This is a single hosted MCP (Model Context Protocol) server. Point your AI agent at one streamable-HTTP endpoint and it gains 10 live research tools:

• arXiv — search and read academic preprints (CS, physics, math, and more)
• Reddit — search posts and comments across subreddits
• GitHub — search repositories
• Hacker News — search stories and discussion
• OpenAlex — query the open catalog of scholarly works, authors, and citations
• Wikipedia — look up and read articles
• CoinGecko — fetch cryptocurrency prices and market data
• GDELT — search global news coverage across languages and countries
• Telegram — read public channel messages
• Stack Exchange — search Stack Overflow and the wider Stack Exchange network

Instead of registering for ten different APIs, managing ten sets of credentials, and writing ten client wrappers, you add one server. Your agent decides which tool to call based on the question, then grounds its answer in live data it just pulled.

Who it's for

• Developers building AI agents who need grounded retrieval without maintaining a pile of API integrations.
• Claude, Cursor, and Claude Code power users who want their assistant to search real sources mid-conversation instead of guessing from training data.
• Researchers, indie devs, and AI tinkerers who want answers backed by papers, code, and news — without signing up for accounts at arXiv, Reddit, GitHub, and the rest.

In short: it's the layer that lets your agent go look something up instead of answering from training data.

Quick start — how to connect

The server speaks streamable HTTP, the current MCP transport. Your endpoint is:

https://dami-studio--research-mcp-server.apify.actor/mcp?token=<APIFY_API_TOKEN>

Replace <APIFY_API_TOKEN> with your token from Apify Console → Settings → API & Integrations. The first request cold-starts the server (about 10–20 seconds); calls after that are warm.

Cursor / Claude Code (native HTTP transport)

These clients support remote MCP servers directly. Add to your MCP config:

{
  "mcpServers": {
    "research": {
      "url": "https://dami-studio--research-mcp-server.apify.actor/mcp?token=YOUR_APIFY_API_TOKEN"
    }
  }
}

Claude Desktop (via mcp-remote bridge)

Claude Desktop's config validates local stdio servers, so bridge the HTTP endpoint with mcp-remote. Edit claude_desktop_config.json (%APPDATA%\Claude\ on Windows, ~/Library/Application Support/Claude/ on macOS):

{
  "mcpServers": {
    "research": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://dami-studio--research-mcp-server.apify.actor/mcp?token=YOUR_APIFY_API_TOKEN"
      ]
    }
  }
}

Fully quit and reopen Claude Desktop after saving. You can verify the connection with an MCP initialize handshake — the server returns serverInfo.name = research-mcp-server, version 1.0.0, and capabilities.tools.

You can also add this server through the Apify MCP server at mcp.apify.com.

Input

This is an MCP server, so you don't fill in a normal Actor input form for everyday use — your agent calls tools at runtime. If you run it directly (for example, to keep it warm or test it), the relevant settings are:

• token — your Apify API token, passed as the ?token= query parameter on the endpoint. Required for authentication.
• standby — the server runs in Apify Standby mode (already enabled), so it stays reachable at a stable URL and accepts MCP requests on demand.

Each underlying tool takes its own arguments at call time — for example, a search query string, a subreddit name, or a result limit — which your MCP client passes through automatically.

Output

Tools return structured JSON that the MCP client hands to your agent as tool results. The shape depends on the tool, but a typical result row looks like:

• arXiv paper — title, authors, summary, published date, arxiv_id, and a link to the PDF.
• Reddit post — title, subreddit, author, score, num_comments, created timestamp, and permalink.
• GitHub repo — full_name, description, stars, language, url, and updated_at.
• CoinGecko price — id, symbol, current_price, market_cap, and price_change_percentage_24h.

Results are designed to be readable by an LLM directly: clear field names, source links so the model can cite, and timestamps so it can reason about recency.

Example use cases

• See all 10 tools served from one endpoint
• Search arXiv for recent papers from inside your AI agent
• Search a subreddit for real user discussion as an MCP tool
• Search GitHub repositories from your assistant

Pricing

There is no per-result charge on this Actor — it's effectively free to use. You pay only standard Apify platform usage (the compute the server consumes while running). The underlying sources — arXiv, Reddit, GitHub, Hacker News, OpenAlex, Wikipedia, CoinGecko, GDELT, Telegram, and Stack Exchange — need no API keys of your own.

FAQ

Do I need API keys for arXiv, Reddit, GitHub, and the others? No. The server handles all source access. You only need an Apify API token to authenticate to the endpoint.

Which clients work with it? Any MCP client that supports streamable HTTP. Cursor and Claude Code connect via the native url field; Claude Desktop connects through the mcp-remote bridge shown above. ChatGPT and other MCP-capable agents work the same way.

Why is the first request slow? The server runs in Apify Standby and cold-starts on the first call after idle — roughly 10–20 seconds. Subsequent calls hit the warm instance and respond quickly.

Is it really free? Yes, in the sense that the Actor sets no per-result price. You're billed only for the standard Apify platform compute the server uses, which is minimal for typical research queries.

Can my agent use several tools in one conversation? Yes. All 10 tools come from the single endpoint, so the agent can search arXiv, then Reddit, then GitHub within the same session without switching servers or credentials.

How do I confirm it's connected? Run an MCP initialize handshake from your client. A working server reports serverInfo.name = research-mcp-server, version 1.0.0, and advertises capabilities.tools with all 10 tools listed.