Use Ethical Stack with your agent.
Find AI choices, compare the evidence and follow every finding back to its sources.
Start with a question
Try: “Compare coding tools on privacy and training-data rights. Explain the tradeoffs, cite the sources and flag what is still unknown.”
Give your agent the research index, or connect an MCP client to https://ethicalstack.ai/mcp. The service is public and read-only; no account or API key is required.
Read a page as Markdown
Add index.md to a page URL, or request the page with Accept: text/markdown. For example: the scoring methodology as Markdown. Profiles and guides preserve the findings and sources in plain text.
Look up selected research
The API returns small results instead of requiring the full research download. Start with task ids and dimension names at /api/v1/methodology.
/api/v1/profiles?task=harnesses&dimensions=privacy,training-rights&limit=3finds coding tools./api/v1/assessment?key=harnesses/claude-code&dimensions=privacy,training-rightsretrieves cited findings./api/v1/compare?keys=harnesses/claude-code,harnesses/aider&dimensions=privacy,training-rightscompares selected tools.
Read the API specification for filtering, pagination and parameters. Minimum-score filters exclude findings without a numerical score. Results retain the published overall calculation; selecting a dimension does not recalculate it.
Connect through MCP
The Streamable HTTP endpoint exposes search_profiles, get_assessment, compare_profiles and get_methodology. It performs research lookups without changing data or making purchases. Server clients can connect directly; browser clients must use the site origin or an origin explicitly configured by the operator.
Connection metadata and a short research skill help compatible agents discover the service and interpret its evidence. Discovery formats are still evolving.
Keep the evidence limits
Ask for the profile link, primary sources, confidence, scope and unknowns. A company relationship does not prove use of a particular product. Local capability depends on the configuration. Energy scores describe practices and transparency rather than measured emissions. Missing evidence is not proof of good or bad conduct. BDS findings are limited to the campaign sources reviewed.
Source-check and review dates describe the research; an API lookup does not reverify the source. Read how the scores work.
Content use and privacy
Our Content Signals allow search and AI input while declining AI training. These state preferences; they do not enforce access or replace third-party source rights. API and MCP parameters are processed by the server. No analytics events are sent for those requests; Cloudflare’s hosting practices still apply. Avoid sending private information in a query.