APITube News

APITube News MCP

Real-time and archived news search for AI agents, over the Model Context Protocol. Hosted at https://mcp.apitube.io/ — no package to install, no local process to keep alive.

Quick Start • Tools • Filters • Prompts • Pricing • Troubleshooting • Docs • Support

300,000+ sources · 177 countries · 59 languages. Sentiment and entities on every article.

---

Overview

The APITube MCP server gives an assistant live access to the world's news as structured data, not scraped HTML. It exposes 2 tools:

• `search_news` — most of the News API filter set in one call: keywords, language, country, source domain and quality rank, sentiment range, named entities, media, date ranges, sorting, faceting and highlighting.
• `suggest` — resolves a name like "Tesla" into the entity, category, topic and industry IDs the precise filters need.

Every article comes back enriched by the pipeline behind it: sentiment scores, extracted entities (people, organizations, locations, brands, events), IPTC categories, topics and industries.

MCP client  →  mcp.apitube.io  →  api.apitube.io

(this server) (News API)

JSON-RPC over HTTP, Authorization: Bearer

Quick Start

1. Get an API key at apitube.io.
2. Add the server to your client with the block below — each one is also a ready file in configs/.
3. Restart the client and ask it something like *"find positive breaking news about Tesla in English from the last week"*.

Claude Code

claude mcp add --transport http apitube-news https://mcp.apitube.io/ \

--header "Authorization: Bearer YOUR_API_KEY"

Check it with /mcp. To commit the server to a project instead, put configs/claude-code.mcp.json at the repo root as .mcp.json.

Cline

MCP Servers → Configure, or ~/.cline/mcp.json for the CLI:

{

"mcpServers": { "apitube-news": { "type": "streamableHttp", "url": "https://mcp.apitube.io/", "headers": { "Authorization": "Bearer YOUR_API_KEY" }, "disabled": false, "autoApprove": [] } } }

type must be set explicitly — without it Cline falls back to the legacy SSE transport, which this server does not serve. Both tools are read-only, so autoApprove: ["search_news", "suggest"] is safe if you would rather not confirm every call.

Cursor

~/.cursor/mcp.json (global) or .cursor/mcp.json (per project):

{

"mcpServers": { "apitube-news": { "url": "https://mcp.apitube.io/", "headers": { "Authorization": "Bearer YOUR_API_KEY" } } } }

Settings → MCP should list apitube-news as connected.

Claude Desktop

Claude Desktop only launches local processes, so bridge the hosted server with mcp-remote. Edit claude_desktop_config.json (~/Library/Application Support/Claude/ on macOS, %APPDATA%\Claude\ on Windows):

{

"mcpServers": { "apitube-news": { "command": "npx", "args": [ "-y", "mcp-remote", "https://mcp.apitube.io/", "--header", "Authorization: Bearer YOUR_API_KEY" ] } } }

Restart the app; the tools appear under the slider icon.

VS Code (GitHub Copilot)

.vscode/mcp.json, with the key prompted instead of stored in plain text:

{

"inputs": [ { "type": "promptString", "id": "apitube-key", "description": "APITube API Key", "password": true } ], "servers": { "apitube-news": { "type": "http", "url": "https://mcp.apitube.io/", "headers": { "Authorization": "Bearer ${input:apitube-key}" } } } }

Open Copilot Chat in Agent mode and enable the apitube-news tools.

Windsurf

~/.codeium/windsurf/mcp_config.json — note serverUrl, not url:

{

"mcpServers": { "apitube-news": { "serverUrl": "https://mcp.apitube.io/", "headers": { "Authorization": "Bearer YOUR_API_KEY" } } } }

Windsurf Settings → Cascade → MCP Servers → refresh.

Any other client, or plain curl

Anything that speaks Streamable HTTP takes the URL directly; clients limited to stdio go through mcp-remote, as in the Claude Desktop block. The handshake needs no key:

curl -s -X POST https://mcp.apitube.io/ \

-H "Content-Type: application/json" \ -d '{"jsonrpc":"2.0","method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"curl","version":"1.0.0"}},"id":1}'

{

"jsonrpc": "2.0", "result": { "protocolVersion": "2024-11-05", "serverInfo": { "name": "APITube News MCP-Server", "version": "1.0.0" }, "capabilities": { "tools": { "listChanged": true }, "prompts": { "listChanged": true } } } }

A real search adds the key:

curl -s -X POST https://mcp.apitube.io/ \

-H "Content-Type: application/json" \ -H "Authorization: Bearer YOUR_API_KEY" \ -d '{"jsonrpc":"2.0","method":"tools/call","params":{"name":"search_news","arguments":{"title":"Bitcoin","language":{"code":"en"},"per_page":5}},"id":1}'

Tools

search_news

Arguments are nested objects, never dotted strings:

{ "language": { "code": "en" } }      // ✅

{ "language.code": "en" } // ❌ rejected

Three things worth knowing before the first call:

• The article body is not returned by default. The default field list is id,title,href,published_at,description,source.domain. Ask for the text explicitly with fl: "title,href,body".
• One response carries at most 25 articles. per_page defaults to 10 and is clamped to 25; walk further with page. If a result set still has to be cut — 25 articles of full text can be large — the response gains an _mcp_truncated field saying how many were omitted.
• A title search spans at most 31 days. With no dates it covers the last 31 days; a wider explicit range fails with 400 ER0110. Split longer periods into month-sized windows. Searches without a title filter have no range limit.

Misspelled arguments are rejected with JSON-RPC -32602 and a suggestion, instead of being silently ignored:

Unknown parameter 'langauge.code'. Did you mean 'language.code'?

export, query and prompt are deliberately not exposed.

suggest

The precise filters take IDs you cannot guess, so resolve them first:

suggest({ type: "entities", prefix: "Tesla" })

// → [{ id: 474, name: "Tesla Robotaxi", type: "brand", … }, …]

search_news({ entity: { id: "474" }, language: { code: "en" } })

type is one of entities, categories, topics, industries; prefix is a name or its beginning. Both are required. Matching is by prefix, so read the names before filtering on the first hit.

Filters

Everything below belongs to search_news. Content, taxonomy, language, author and source filters have an ignore.* twin for exclusion (ignore.title, ignore.entity.id, ignore.source.domain, …); sentiment, media and time filters do not. Multi-value filters take up to 3 comma-separated values. has_* and is_* take 0 or 1, not true/false.

Content and taxonomy

Sentiment

Sources and quality

Media, shape and time

Output: sorting, paging, faceting, highlighting

| fl | `"id,title,source.name,sentim