The Biel.ai MCP Server connects your IDE to product documentation, allowing AI tools such as Cursor, VS Code, and Claude Desktop to access a company's product knowledge through the Biel.ai platform.
Purpose Biel.ai provides a hosted retrieval-augmented generation (RAG) layer that makes your documentation searchable and usable by AI tools. This enables smarter completions, accurate technical answers, and context-aware suggestions in the IDE or chat environment. When AI tools can read product documentation, they become more helpful: they generate more accurate code completions, answer technical questions with context, and guide developers with real-time product knowledge.
Main capabilities The server exposes three tools: biel_ai: lets Biel.ai generate answers with conversational context; existing calls using message continue to generate answers. biel_search: performs hybrid keyword and semantic search over indexed web pages, uploaded files, repositories, and OpenAPI sources, without generating answers. biel_get_document: reads the full indexed text of a document returned by search.
Usage
- Get the MCP configuration: required project_slug and domain; optional api_key (only needed for private projects). A Biel.ai account and project setup are required; an official 15-day free trial is available.
- Add it to your AI tool: in Cursor, add it under Settings → Tools & Integrations → New MCP server; in Claude Desktop, edit claude_desktop_config.json; in VS Code, install the MCP extension.
- Start asking questions.
When calling biel_search, you can use natural language questions or search terms. The search request API uses search_type=hybrid&source_types=all; the REST API defaults to search_type=keyword, regardless of the selected sources. Results include the chunk text, source type, document_id, and page/worksheet references; non-URL documents are referenced by title and location. The limit parameter controls the number of returned chunks (default 5, maximum 20).
Using biel_get_document And pass in the document_id of a certain search result to read the complete indexed document without accessing its source site. get returns the complete indexed chunks in reading order, with PDF page numbers and spreadsheet worksheet references. Large documents will include a next_cursor; call repeatedly with that cursor and the same document_id and limit until next_cursor no longer exists. It returns indexed text, not the original binary uploaded content or live page HTML. References belong to the project's currently active index generation: after re-crawling, references become invalid and you need to search again. Older repository and OpenAPI sources need to be re-crawled to populate the parent document identifier before get can be used.
Search and get do not create or continue a chat. Private projects require an API key with the search scope; generating answers uses chats_create. All operations perform the API's project access, domain, and quota checks. The default workflow is search → get → answer, using the retrieved text. If the calling agent wants Biel.ai to generate the answer, it should first ask the user, unless the user has already explicitly requested it. A retrieval failure does not automatically switch to generating an answer.
Self-hosting (optional) For advanced users who want to run their own MCP server instance: Local development. Use as a Python package: applications hosting their own ASGI stack can install the server directly from a tagged release and import its FastAPI app. Docker deployment.
Use cases Suitable for developers querying company product documentation in an IDE or chat tool, obtaining technical Q&A, code completion suggestions, and retrieving product knowledge via RAG.
Limitations and support Requires a Biel.ai account and project setup (officially offers a 15-day free trial). Private projects require an API key and the corresponding scope. Retrieval references become invalid due to re-crawling, and older repository and OpenAPI sources must be re-crawled before get can be used. A retrieval failure does not automatically generate an answer. Feedback can be provided via GitHub Issues; contact email support@biel.ai; custom demos can be booked.
