Measures independent news coverage with syndication-aware source counting, confidence scores, evidence, and coverage caveats.
Copy the AI prompt to install this server into Claude Code, Cursor, or another agent โ or use 1-click editor setup below.
This server is confirmed live โ we successfully called its tools/list endpoint directly (see the verified badge above). We haven't yet sandbox-tested the stdio install command below specifically, which is a separate, ongoing check.
๐ก Paste the JSON block into your client's configuration file under mcpServers, then restart the application.
Inspect callable tools, capabilities, and parameters exposed to AI agents by Corroborate MCP.
corroborate_claimGiven a factual/news claim, returns how INDEPENDENTLY it is being reported. Use before relying on a current-event claim. Output: verdict (CONFIRMED = 2+ independent origins | SINGLE_SOURCE | UNCORROBORATED), count of independent story origins (wire/syndication echoes collapse to one), per-source evidence (outlet, domain, url, date), confidence 0-1, coverage flag, honest caveats. Measures REPORTING corroboration, not truth โ no stance detection, so a distorted claim about a real event may still show coverage; verify specifics against the returned sources. Keyless, read-only, deterministic; if all sources are unreachable it errors rather than returning a false negative.
find_sourcesMulti-engine news/source search (Google News, GDELT, Hacker News) for a query: a deduped list of {outlet, domain, url, date}. Use when you want raw coverage to judge yourself, not a scored verdict โ the cheaper primitive under corroborate_claim. Keyless, read-only.
The chefcohen/corroborate-mcp MCP server measures reporting corroboration rather than truth. Give it a factual or news claim, and it searches multiple public sources to estimate how many independent story origins address that claim. Syndicated or near-identical reports are grouped so that many copies of one wire story do not appear to be many independent confirmations.
The main result uses three verdicts:
CONFIRMED: at least two independent story origins were found.SINGLE_SOURCE: one independent origin was found.UNCORROBORATED: no qualifying origin was found.Results include the independent-source count, confidence from 0 to 1, coverage status, source details, and notes about limitations such as syndication, relevance filtering, or engine outages. A confirmed result means that the topic has independent reporting; it is not a factual guarantee.
The chefcohen/corroborate-mcp MCP server searches Google News RSS, GDELT, and Hacker News in parallel. It extracts keywords from the claim while preserving quoted phrases, then applies a relevance check to avoid counting articles that only loosely relate to the subject.
Relevant results are clustered by headline similarity. Domains carrying substantially the same story can therefore count as one origin, while distinct reporting may count separately. Wire-service origins are identified in the source evidence. Confidence increases with the number of origins and agreement between search engines, and can decrease when a single story is echoed widely.
The default recency window is seven days, and callers can request a different window up to 90 days. Because the process does not use an LLM, identical evidence produces deterministic judgments.
Node.js 18 or newer is required. No API keys, accounts, or configuration values are needed. Start the server over stdio with:
An MCP client can register npx as the command, with -y and corroborate-mcp as its arguments. The README provides configurations for Claude Desktop and Cursor, and also shows a Claude Code registration command. The package can be run without an MCP client for a basic stdio health check.
corroborate_claim accepts a claim plus optional window_days and max_sources values. It returns the verdict, independent-origin count, confidence, coverage state, per-source outlet and domain information, URLs, publication dates, wire indicators, and explanatory notes. If every search engine is unreachable, it returns an error instead of treating missing results as evidence that the claim is unsupported.
find_sources accepts a query with optional recency and source-count limits. It returns a deduplicated list of outlets, domains, URLs, and dates without assigning a verdict. Use it when an agent needs to inspect the available coverage directly or apply its own judgment.
The chefcohen/corroborate-mcp MCP server does not perform stance detection. A distorted claim about an event that is genuinely being reported can still receive CONFIRMED, so returned sources must be checked for the exact details. It works at the English-language, headline level and does not fetch paywalled article bodies.
Results are constrained by the selected time window. An old claim may be labeled UNCORROBORATED simply because its coverage falls outside that window. Independent rewrites can occasionally be grouped incorrectly or counted separately, and search-engine outages can reduce coverage. The project reports published benchmark results and identifies these weaknesses rather than presenting corroboration as proof of truth.
Always refer to the official documentation for the most accurate and up-to-date information.
Factual signals from GitHub, npm, and our automated checks โ not a rating.
No reviews yet โ be the first to share how this listing worked for you.
Showcase your server listing on GitHub or your project documentation. Embed this dynamic SVG badge to highlight official listing status and live engagement.
[](https://allmcps.com/mcp/chefcohen-corroborate-mcp)<a href="https://allmcps.com/mcp/chefcohen-corroborate-mcp"><img src="https://allmcps.com/api/badge/chefcohen-corroborate-mcp?style=directory" alt="Corroborate MCP on AllMCPs" /></a>