# klaro.legal MCP server: contracts and official letters in plain language Paste a contract, an official letter, a tax assessment or a set of terms and conditions, and get it back explained clause by clause in everyday language, from Claude, ChatGPT, Cursor or any MCP client. Text, a PDF or photos of the pages all work: a photographed letter is read by OCR before the analysis. Six languages. No account, no API key. Hosted by [klaro.legal](https://klaro.legal); this repository holds the public metadata, the server code is not open source. ``` https://klaro.legal/api/mcp ``` Streamable HTTP · no authentication · listed in the [official MCP registry](https://registry.modelcontextprotocol.io/v0/servers?search=klaro) as `legal.klaro/document-explainer` and as a [Glama connector](https://glama.ai/mcp/connectors/legal.klaro/document-explainer). [](https://glama.ai/mcp/connectors/legal.klaro/document-explainer) ## What it explains, and what it does not It explains **what a document says**. It does not review it, score it, flag clauses as unfair, or tell you what to do. That line is deliberate: klaro.legal is software, not a law firm. What matters in your case is for a qualified lawyer to decide, and every answer says so. Typical documents: rental agreements, employment contracts, letters from a tax office or an immigration authority, insurance policies, terms and conditions, utility and telecoms contracts. ## Tools - `explain_document`: send the document as `text`, as a base64-encoded `pdf`, or as base64-encoded `images` of its pages (JPEG, PNG, HEIC, WebP; up to 15 pages and 8 MB decoded). Exactly one of the three. `language` picks the language of the explanation. Returns a short summary plus one entry per clause: the original wording, then what it means. Plain text takes 20 to 60 seconds, a scan or photo longer because of OCR. *read-only* - `get_explanation`: if `explain_document` was still running, it returns an `analysis_id` — pass it here to pick up the finished result. *read-only* Explanation languages: `de`, `en`, `es`, `fr`, `it`, `pt`. The document itself may be in any language. ## Setup ```json { "mcpServers": { "klaro-legal": { "type": "streamableHttp", "url": "https://klaro.legal/api/mcp" } } } ``` See [llms-install.md](llms-install.md) for a one-call check that it works. ## Text, PDF or photos Three ways in, one at a time: - **`text`** — the fastest route. Most clients already extract the text of a PDF by themselves, and then this is all that is needed. - **`pdf`** — the file itself, base64-encoded, up to 8 MB decoded. Used when the PDF has no usable text layer, or when its layout carries meaning. - **`images`** — photos or scans of the pages, base64-encoded, in page order, up to 15 pages and 8 MB decoded in total. This is the route for a letter someone photographed with a phone: OCR reads it first, then the analysis runs. Anything larger than 8 MB is better handled by the [upload form](https://klaro.legal/en-us), which takes files up to 50 MB — a base64 payload travels through the client's context window, and that is where the limit comes from, not from the analysis. ## Privacy Documents are anonymised before the analysis and processed inside the EU. The analysis itself runs on AWS Bedrock in Paris (`eu-west-3`); the text is not sent to any other AI provider. Details: [klaro.legal/en-us/privacy](https://klaro.legal/en-us/privacy) ## Also available - **Upload form** in the browser, for files above 8 MB and up to 50 MB: [klaro.legal](https://klaro.legal/en-us) - **Embeddable widget** for your own site, in all six interface languages: [klaro.legal/en-us/embed-widget](https://klaro.legal/en-us/embed-widget) --- Operated by Martin Schenk S.L., Madrid · [klaro.legal](https://klaro.legal) · questions: info@klaro.legal