#
MCP Server

The AskNews MCP Server connects AI assistants like Claude and ChatGPT to AskNews DeepNews research capabilities using the Model Context Protocol. Give your AI access to real-time news research, fact-checking, and forecasting.

#
What is MCP?

MCP (Model Context Protocol) is an open standard created by Anthropic for connecting AI applications to external tools and data sources. Think of it like a USB-C port for AI - it provides a standardized way to connect AI assistants to the systems they need.

With the AskNews MCP server, your AI assistant can:

Deep Research (do_news_research): Perform iterative research on complex questions with fast or rich analysis modes
News Search (search_news): Search real-time news with granular filters (categories, sentiment, entities, languages, countries)
Wikipedia (search_wikipedia): Search Wikipedia content with semantic search and contextual chunks
Web Search (search_web): Live web search with LLM-distilled, structured responses
X/Twitter (search_x_twitter): Search Twitter/X using advanced query syntax for social monitoring

#
Requirements

AI assistant with MCP support: Claude (Pro, Max, Team, or Enterprise) or ChatGPT (Plus, Pro, Business, Enterprise, or Education with Developer Mode)
AskNews account: With API access (sign up here)

#
Programmatic Usage

Beyond interactive AI assistant sessions, you can use the AskNews MCP server programmatically in your own applications. This is useful for building AI agents, automation pipelines, prediction systems, and custom integrations.

#
Getting an API Key

For programmatic access, you'll need an API key:

Go to https://my.asknews.app/en/settings/api-credentials
Log in to your AskNews account
Click Create API Key
Select the required scopes: chat, news, stories
Copy and securely store your API key

#
Using MCP Tools with LangChain

If you're using LangChain, you can easily integrate AskNews MCP tools using the langchain-mcp-adapters library. This allows LangChain agents to use MCP tools directly.

Install the adapter:

Use AskNews MCP tools in your agent:

Filtering specific tools:

You can filter which tools to load if you only need specific functionality:

Benefits of using LangChain MCP adapters:

Seamlessly integrate AskNews with other LangChain tools
Use with any LangChain agent (ReAct, function calling, etc.)
Automatic tool discovery and schema validation
Support for both stateless and stateful interactions

#
Claude Code CLI

For developers using Claude Code, you can add the MCP server directly:

Option 1: OAuth Authentication (Recommended)

After adding, authenticate using the /mcp command in Claude Code:

> /mcp
# Select "Authenticate" for AskNews and complete the browser login

Option 2: API Key Authentication

Managing the server:

#
Building a Prediction Agent

Here's a complete example of building an AI agent that uses AskNews MCP for making predictions. This example uses langchain-mcp-adapters for seamless integration with LangChain agents:

#
Direct Client Usage

Install the MCP SDK:

Connect to AskNews MCP and call the do_news_research tool:

#
Project Configuration (.mcp.json)

For team projects using Claude Code, share the MCP configuration via a .mcp.json file:

{
  "mcpServers": {
    "asknews": {
      "type": "http",
      "url": "https://mcp.asknews.app",
      "headers": {
        "X-API-Key": "${ASKNEWS_API_KEY}"
      }
    }
  }
}

Team members set the ASKNEWS_API_KEY environment variable locally, keeping credentials secure while sharing the configuration.

#
Connecting to Claude

#
Claude.ai (Web)

Navigate to Settings > Connectors at claude.ai/settings/connectors
Click "Add custom connector" at the bottom of the Connectors section
Enter the server URL: https://mcp.asknews.app
Click "Add"
Click "Connect" to begin authentication
Sign in with your AskNews credentials and grant the requested permissions

#
Claude Desktop

You can add the AskNews connector directly in Claude Desktop or via the web interface (connectors sync across both):

Option 1: Add in Claude Desktop

Open Claude Desktop
Navigate to Settings > Connectors
Click "Add custom connector" at the bottom of the Connectors section
Enter the server URL: https://mcp.asknews.app
Click "Add"
Click "Connect" to begin authentication
Sign in with your AskNews credentials and grant the requested permissions

Option 2: Add via Web (syncs automatically)

Add the connector at claude.ai/settings/connectors (steps above)
Open or restart Claude Desktop - the connector will automatically sync from your account

Once configured, access tools via the "Search and tools" button in your chat.

#
Claude Mobile (iOS/Android)

Claude Mobile uses connectors already configured on your claude.ai account. Add the connector via the web interface first, then the tools will be available on mobile.

#
Team & Enterprise Plans

For Team and Enterprise plans, Owners or Primary Owners must first enable the connector:

Navigate to Admin settings > Connectors
Click "Add custom connector"
Enter the server URL: https://mcp.asknews.app
Click "Add"

Individual users then connect and authenticate with their own AskNews credentials. This ensures Claude only accesses data each user has permission to use.

#
Connecting to ChatGPT

ChatGPT supports MCP servers through Developer Mode (available on Plus, Pro, Business, Enterprise, and Education plans).

#
ChatGPT Web

Navigate to Settings in ChatGPT
Go to Apps & Connectors > Advanced settings
Enable Developer mode
Return to Apps & Connectors and click "Create" (now enabled with Developer mode)
Enter the server URL: https://mcp.asknews.app
Complete the authentication flow with your AskNews credentials

#
Notes on ChatGPT MCP Support

MCP support in ChatGPT is currently in beta
Some MCP features may have limited compatibility compared to Claude
For the most stable experience, we recommend using Claude for MCP integrations
Check OpenAI's MCP documentation for the latest updates

#
Connecting to HuggingFace Chat

HuggingFace Chat supports MCP servers, allowing you to enhance your conversations with real-time news research capabilities.

#
HuggingFace Chat Web

Navigate to HuggingFace Chat
Click "MCP Servers" in the left menu
A modal will appear - click "Add Server"
Enter the server URL: https://mcp.asknews.app
Add the authentication header:
- Header name: X-API-Key
- Header value: Your AskNews API key (get one from API credentials)
Click "Add" to save the server configuration

Once configured, enable the server using the switch button in the custom servers section. After enabling, the AskNews tools will be available in your chat sessions and the assistant can automatically invoke them when answering questions that require current news information.

#
Using Tools in HuggingFace Chat

Tools are automatically invoked when relevant to your query
You can see which tools were used in the conversation thread
Tool results are seamlessly integrated into the assistant's responses

#
Notes on HuggingFace Chat MCP Support

HuggingFace Chat uses API key authentication for MCP servers
Tools are invoked automatically based on conversation context
For questions or issues, visit the HuggingFace Chat documentation

#
Authentication

The AskNews MCP server uses OAuth2 authentication. When you click "Connect":

You'll be redirected to AskNews login
Sign in with your AskNews credentials
Grant permission to access your account
You'll be redirected back to your AI assistant

Your access token is securely stored and automatically refreshed. You can revoke access at any time by disconnecting the connector in your AI assistant's settings or in your AskNews account settings.

#
Available Tools

Once connected, enable tools via the "Search and tools" menu in your AI assistant's chat interface.

#
do_news_research

State-of-the-art news search agent capable of retrieving in-depth news across all languages, countries, and sources. Excellent for forecasting, fact-checking, and complex research questions that require iterative investigation.

Parameters:

Parameter	Type	Default	Description
`query`	string	required	The research question or news topic. Include timeframe if relevant (e.g., "in the last 24 hours", "this week", "during 2024"). Specify keywords and optionally where to find information.
`model`	`"fast"` or `"rich"`	`"fast"`	`"fast"` for quick shallow retrieval when keywords are known. `"rich"` for in-depth iterative research when keywords aren't clear (takes longer).
`sources`	array of strings	`["asknews"]`	List of data sources: `asknews`, `google`, `wiki`, `x`. Order determines prioritization.

Response:

Returns a DeepNewsResponse containing:

overview: Research summary with numbered inline citations [1], [2], etc.
sources: List of source objects with origin, id, title, summary, URL, publication date, and source

Important: When your AI assistant presents results, it preserves the numbered citations from the overview. These reference the sources list for verification.

#
search_news

Search for enriched real-time news context from the last 5 minutes up to 160 days in the historical archive. Optimized for low-latency applications with extensive filtering options.

Parameters:

Parameter	Type	Default	Description
`query`	string	`""`	Semantic search query. Natural language supported, keywords prioritized. Leave blank for filter-only search.
`n_articles`	integer	`10`	Number of articles to return (usually 10-30 depending on plan).
`hours_back`	integer	`24`	Hours to look back (up to 3840 hours / 160 days).
`categories`	string	`null`	Filter by categories: Business, Crime, Politics, Science, Sports, Technology, Military, Health, Entertainment, Finance, Culture, Climate, etc.
`languages`	string	`null`	Two-letter ISO 639-1 codes (e.g., "en", "de", "es", "fr").
`countries`	string	`null`	Two-letter ISO country codes for publisher location (e.g., "US", "FR", "SE").
`domain_url`	string	`null`	Filter by specific domains (e.g., "npr.org", "nature.com").
`string_guarantee`	string	`null`	Comma-separated strings that must appear in results.
`entity_guarantee`	string	`null`	Entity filters like "Person:John Smith", "Location:Paris", "Organization:OpenAI".
`sentiment`	`"negative"`, `"neutral"`, `"positive"`	`null`	Filter by article sentiment.
`reporting_voice`	string	`null`	Filter by voice: Objective, Investigative, Analytical, Narrative, etc.
`return_type`	`"dicts"`, `"string"`, `"both"`	`"string"`	Output format. "string" is prompt-optimized, "dicts" includes full metadata.

Response:

Returns a SearchResponse containing:

as_string: Prompt-optimized string format (if return_type is "string" or "both")
as_dicts: List of article dictionaries with full metadata including summaries, sentiment, entities, reporting voice, and more (if return_type is "dicts" or "both")
offset: Pagination offset for subsequent queries

#
search_wikipedia

Search Wikipedia content using natural language. Find relevant chunks with contextual neighbor chunks and full articles.

Parameters:

Parameter	Type	Default	Description
`query`	string	required	Natural language query - any phrase, keyword, question, or paragraph.
`n_documents`	integer	`5`	Number of documents to return.
`neighbor_chunks`	integer	`1`	Number of contextual neighbor chunks to include with each result.
`full_articles`	boolean	`false`	Return full articles instead of just chunks. Increases token usage.
`hybrid_search`	boolean	`false`	Use hybrid search (vector + keyword) instead of vector-only search.
`string_guarantee`	string	`null`	Comma-separated strings that must appear in results.
`diversify`	float	`0`	Diversity factor for MMR re-ranking (0.0 = pure relevance, 1.0 = full diversity).
`include_main_section`	boolean	`false`	Include the main section of each article for additional context.

Response:

Returns a WikiSearchResponse containing a list of documents with:

content: Text content from the Wikipedia article
title: Article title
url: URL to the Wikipedia article
categories: List of Wikipedia categories
timestamp: Last update time

#
search_web

Run a live web search and get back a fully structured, LLM-distilled response. Perfect for up-to-date information with intelligent summarization.

Parameters:

Parameter	Type	Default	Description
`query`	string	required	Keyword-focused query optimized for Google-style web search.
`lookback`	integer	`null`	Hours to look back. Defaults to all time if not specified.
`domains`	string	`null`	Comma-separated domains to search (e.g., "wikipedia.org,reddit.com").
`strict`	boolean	`false`	Only return results with known publication dates within the lookback period.
`offset`	integer	`0`	Pagination offset for subsequent queries.
`return_type`	`"dicts"`, `"string"`, `"both"`	`"string"`	Output format. "string" is prompt-optimized, "dicts" includes full metadata.

Response:

Returns a WebSearchResponse containing:

as_string: Prompt-optimized LLM distillation ready for injection into prompts
as_dicts: List of result objects with title, url, source, published date, key_points, and raw_text
offset: Pagination offset for more results

#
search_x_twitter

Search Twitter/X using Twitter Advanced Search query language. Perfect for finding specific tweets, tracking conversations, and monitoring sentiment.

Parameters:

Parameter	Type	Default	Description
`query`	string	required	Twitter Advanced Search query. Supports: `or`, `and`, `from:username`, `@username`, `since:YYYY-MM-DD_HH:MM:SS_UTC`, `until:YYYY-MM-DD_HH:MM:SS_UTC`, `#hashtag`, `filter:links`, `filter:media`, `min_faves:N`, etc.
`lookback`	integer	`null`	Hours to look back. Use instead of since/until for relative time filtering.
`strict`	boolean	`false`	Only return results with known publication dates within the lookback period.
`offset`	integer	`0`	Pagination offset for subsequent queries.
`return_type`	`"dicts"`, `"string"`, `"both"`	`"string"`	Output format. "string" is prompt-optimized, "dicts" includes full metadata.

Example queries:

# Search for Bitcoin discussions from a specific user
from:elonmusk bitcoin

# Search for AI news with engagement filters
(artificial intelligence or AI) min_faves:100

# Complex query with multiple operators
(bitcoin or ethereum) and (crash or rally) since:2025-01-01_00:00:00_UTC

Response:

Returns a WebSearchResponse containing:

as_string: Prompt-optimized LLM distillation ready for injection into prompts
as_dicts: List of result objects with title, url, source, published date, key_points, and raw_text
offset: Pagination offset for more results

#
Example Prompts

Here are examples of how to use the AskNews connector. Make sure the AskNews tools are enabled via the "Search and tools" menu before trying these prompts:

#
Deep Research (do_news_research)

Quick news retrieval:

Use the do_news_research tool to find AI news from the last 24 hours

Deep research with rich model:

Use do_news_research with the rich model to research what experts are predicting for the upcoming Federal Reserve meeting

Multi-source research:

Research Tesla's stock performance this week using do_news_research with asknews, google, and x sources

#
News Search (search_news)

Filtered news search:

Search for technology news about Apple from US sources in the last 48 hours

Category-based search:

Find Finance news with positive sentiment from the last week

Entity-based search:

Search for news mentioning Person:Elon Musk and Organization:Tesla from the last 72 hours

#
Wikipedia (search_wikipedia)

Topic research:

Search Wikipedia for information about quantum computing and its applications

Deep article search:

Find Wikipedia articles about the history of artificial intelligence with full article content

#
Web Search (search_web)

General web search:

Search the web for recent OpenAI product announcements

Domain-specific search:

Search reddit.com and hackernews for discussions about the new Claude model

#
Twitter/X (search_x_twitter)

User-specific search:

Search X for tweets from @elonmusk about AI in the last week

Trending topic search:

Search Twitter for discussions about #Bitcoin with high engagement (min_faves:100)

Sentiment monitoring:

Find tweets about the Federal Reserve from the last 48 hours

#
Tips for Better Results

Explicitly mention the tool name in your prompt for precise control
Enable only the tools you need in the "Search and tools" menu
For do_news_research, specify "fast" for quick lookups or "rich" for complex analysis
Use search_news when you need granular filtering (categories, sentiment, entities)
Use search_x_twitter with Twitter's advanced query syntax for targeted social monitoring
Combine tools in your research workflow - start with do_news_research for overview, then drill down with specific tools

#
Using with Claude's Research Mode

When using Claude's Research feature, AskNews tools can be invoked automatically without additional approval. This enables deep, multi-step research across hundreds of sources.

Tips for Research mode:

Only enable tools relevant to your research task
All AskNews tools are read-only and safe to use with Research
Be mindful that Research may make multiple tool calls
do_news_research with the "rich" model is particularly powerful for complex research questions
Combine search_news for recent events with search_wikipedia for background context

#
Security Considerations

Only connect to trusted servers: The official AskNews MCP server URL is https://mcp.asknews.app
Review tool requests: Before clicking "Allow always," review what actions your AI assistant is requesting
Monitor usage: Tool calls use your AskNews API quota
Disconnect anytime: Remove the connector in Settings > Connectors if needed

#
Troubleshooting

#
Connection Issues

Problem	Solution
Can't connect	Verify you have an active AskNews subscription with API access
Authentication failed	Ensure your AskNews account has the required scopes (chat, news, stories)
Connector not appearing	Try disconnecting and reconnecting; restart Claude Desktop

#
Tool Not Working

Problem	Solution
Tool not responding	Verify the tool is enabled in "Search and tools" menu
No results returned	Try a simpler query first to test the connection
Authentication expired	Click "Connect" again to re-authenticate

#
Getting Help

If you continue to experience issues:

Join our Discord for support:
Contact support at contact@asknews.app

DeepNews Documentation - Learn more about DeepNews capabilities via the SDK
Model Context Protocol - Official MCP specification
MCP Python SDK - Official Python client
MCP TypeScript SDK - Official TypeScript client
Claude Connectors Guide - Anthropic's guide to custom connectors
ChatGPT MCP Documentation - OpenAI's guide to MCP connectors
AskNews API Console - Manage your API credentials and usage

Table of Contents

We use cookies