Giving Claude Code access to live news
Add an MCP server that reads news sources, and Claude Code can query it during a session. One command registers the remote Bot Wire server, and every wire answers on the free preview tier immediately. Add a funded Base wallet only when you want full result sets rather than the top three.
/mcp. Nothing here needs an API key or a signup.
Claude Code has no idea what happened this morning. It has file access, a shell, and a model frozen at its training cutoff. News reaches it the same way anything current does: through a tool it calls at the moment it needs one.
The one-line install
claude mcp add --transport http botwire https://thebotwire.com/mcp
Nothing to install and no wallet required. Run /mcp inside a session to confirm the server connected, and you should see five tools: search_news, get_headlines, query_wire, preview_wire and botwire_status.
The remote server returns the free preview tier, which means the top 3 results per query with no summaries. That is enough to answer most questions a coding session actually asks, and enough to judge whether the data is worth paying for.
What Claude Code can usefully ask
The value is not "news" in the general sense. It is the handful of current facts that change how a coding session should proceed.
| You want to know | Wire | Route |
|---|---|---|
| Did a dependency ship a release | releases | /releases/latest?project=node |
| Is there a CVE against what we use | cve | /cve/latest?q=openssl |
| Is the API we are debugging down | status | /status/latest?provider=anthropic |
| What are developers discussing | hn | /hn/latest?feed=frontpage |
| What did a lab just announce | ailab | /ailab/latest?lab=openai |
| General or cross-domain question | news | /news?q=terms&category=tech |
The last row is the fallback, not the default. A domain wire is more precise than a keyword search across general news, and precision costs less because a wasted call still costs a call. The full routing table lists which of the 40 wires owns which question, with the filter enum for each.
Query terms behave like a search index, not a chat prompt. Send Powell inflation rather than a full sentence: this is ranked keyword search, so distinctive nouns beat natural phrasing.
Moving to full results with the local package
The remote server is capped at previews on purpose. For ranked full result sets with summaries, run the local package with a wallet:
claude mcp add botwire -e BOTWIRE_WALLET_PRIVATE_KEY=0x... -- npx -y botwire-mcp
The package signs x402 payments with your key, so calls cost $0.005 to $0.01 in USDC on Base and no API key exists anywhere in the flow. Two practical notes: the key goes into your shell history when typed this way, and a hot wallet holding a few dollars is the right thing to point at a coding agent. Full cost arithmetic, including the tokens each response consumes, is in what a live data query costs an agent.
Why a wire beats pasting links into the session
Pasting a URL asks the model to parse a rendered page, which may be paywalled, JavaScript-dependent, or an article about a primary source rather than the source. A wire response is JSON with published, age_minutes and a relevance score on every item, so the model can judge staleness rather than assume it. And because it is a tool call, retrieval happens when the session needs it instead of when you remembered to paste something.
Freshness is checkable rather than promised. When this page was written, /health reported an observed interval of 279 seconds between ingest runs, with breaking wires polled on every run and sparse government sources throttled to a 15-minute floor. last_poll on that endpoint is the ground truth.
FAQ
Do I need a wallet to try it?
No. The remote MCP server at /mcp needs no wallet, no key and no signup, and every one of the 40 wires answers on it at the free preview tier. A wallet only changes preview results into full ranked results with summaries.
Which wire should Claude Code call before upgrading a dependency?
Two, in order: /releases/latest with the project filter to see whether a new version shipped, then /cve/latest with the package name to check whether the version you are on has an advisory against it. The second one is the reason to bother.
Does the same server work in Claude Desktop and Codex?
Yes, it is the same MCP server. Claude Desktop uses a JSON config file instead of a CLI command, covered in connecting an MCP server to Claude Desktop. Codex registers it with codex mcp add botwire -- npx -y botwire-mcp.
What does Claude Code do if a wire returns nothing?
It sees count:0 and a freshest timestamp, which is information rather than an error. Sparse government and science wires publish daily at most, so the usual fix is a wider since window rather than a different wire.
Sources
Claude Code's MCP configuration, transports and scopes are documented at docs.claude.com. Live wire count, last poll time and observed refresh interval for this service are at thebotwire.com/health.
Related: How do I connect an MCP server to Claude Desktop? · How AI agents get real-time data past the training cutoff · How do AI agents monitor CVEs and security advisories? · How do AI agents check whether AWS or OpenAI is down?