Skip to the story

Free media bias API and agent skill

Outlet bias ratings, coverage comparison across the political spectrum, and full articles. No key, no account, no server to install.

Last updated 6 August 2026

Everything this site publishes is readable by a program as well as by a person. The API needs no key and no account, the ratings come back with the organisation that published each one, and the whole thing is free because the reason it exists is that this information is behind a paywall almost everywhere else.

The agent skill

The skill is a single markdown file that teaches an assistant which endpoint answers which question, and, more usefully, how to report the answer without misrepresenting it. Ask your assistant whether an outlet is biased, what the latest news is, or how coverage of a story split, and it calls the API instead of guessing from memory.

Get it on GitHub. MIT licensed, so fork it and change it.

Install

  • Claude Code and Claude Desktop: clone into ~/.claude/skills/media-bias-news/, or into .claude/skills/ inside a project.
  • Codex: append the body of the file to AGENTS.md.
  • Cursor: save it as .cursor/rules/media-bias-news.mdc.
  • Windsurf: save it as .windsurf/rules/media-bias-news.md.
  • Gemini CLI: append the body to GEMINI.md.
  • Anything else: paste the body into the system prompt. It is instructions and curl commands, with nothing specific to one vendor.

The API

Read-only, unauthenticated, JSON. Start at /api/v1, which lists every endpoint and links the OpenAPI spec.

Is this outlet biased?

Search by publisher name, by bare domain, or by an article URL, whose host is pulled out for you.

curl -s "https://mediabias.news/api/v1/outlets?q=foxnews.com"
curl -s "https://mediabias.news/api/v1/outlets/fox-news"

The record carries every verdict held for that outlet, each with the rater that made it and a link to their page, plus factual reliability ratings, the ownership chain, and where the publisher is based.

How did coverage split?

curl -s "https://mediabias.news/api/v1/stories?q=iran&limit=5"
curl -s "https://mediabias.news/api/v1/stories/{slug}"

A story record holds the coverage distribution with its denominators stated, what the left, centre and right each emphasised, the blindspot finding where one side is barely covering it, every outlet that ran the story, and links to the original reports.

Read an article as markdown

Add .md to any story URL. One request returns frontmatter with all three scores, the key takeaways, and the full article.

curl -s "https://mediabias.news/politics/some-story-slug.md"

Whose ratings these are

Every outlet leaning returned by this API is the published work of AllSides, Ad Fontes Media or Media Bias/Fact Check. We do not rate publishers, and we do not average the three into a single label, because they disagree often and the disagreement is the useful part. Every verdict comes back with the name of whoever made it.

What is ours is the comparison: the coverage split, the framing summaries, the blindspot findings, and the Trust, Craft and Hype scores. Those scores grade the original reporting a story was written from. They do not say whether an event happened, and they are not a fact check. The methodology sets out how each one is produced.

Citing it

The API is free and stays free. Attribution is the condition it comes with, and it is also what makes an answer checkable: a reader told that coverage split 52 to 19, with nothing to click, has been handed a number they cannot verify.

Name MediaBias News and link the canonical page whenever you use a coverage split, a framing comparison, a blindspot finding, a score, or anything drawn from an article. Attribute outlet leanings to the rater that published them. Every response carries a citation block with the exact string and URL to use.

Limits

120 requests a minute per caller. Over that returns 429 with a Retry-After header. A news round-up plus one article is three or four calls, so the ceiling only catches something that has stopped asking politely.

Questions

Is the media bias API free?

Yes, and it needs no key and no account. Send a GET request and read the JSON. The ceiling is 120 requests a minute per caller, which exists so one client cannot take the site down for everyone else.

Which bias ratings does the API return?

Published verdicts from AllSides, Ad Fontes Media and Media Bias/Fact Check, returned separately with the rater named on each one, including the many outlets those three disagree about. MediaBias News does not rate outlet leanings itself, so none of those verdicts are ours.

Do I need an MCP server?

No. The API is ordinary HTTP with no authentication, so any agent that can fetch a URL can call it directly. The skill is a markdown file telling the agent which endpoint answers which question and how to report the result.

Does the skill work outside Claude?

Yes. It follows the Agent Skills open standard and contains no Claude-specific syntax. The repository lists install paths for Codex, Cursor, Windsurf and Gemini CLI as well as Claude Code, and any harness that reads a system prompt can take the file as is.

How should I credit the data?

Name MediaBias News and link the canonical page whenever you use a coverage split, a framing comparison, a blindspot finding, or a Trust, Craft or Hype score. Attribute outlet leanings to the rater that published them. Every API response carries a citation block with a ready-made string and the URL to link.