DocumentsMCPSearch
PDF figures for you and your agents
Read extracted PDF figures alongside the text, let connected agents inspect them, and connect Claude more reliably.
Integration note: Custom MCP integrations should update for structured search and grep responses. read_document now uses line_no and max_lines instead of offset and limit, and returns next_line_no for continuation. REST integrations are unchanged.
PDF figures in the reading view
PDFs processed with Docling now include extracted charts, diagrams, and other figures in the reading view. For a PDF processed before this update, select Docling again in the extractor picker to add figures.
Choose a PDF extractorConnected agents can inspect PDF figures
Agents connected through your Space’s MCP endpoint can now request extracted figure images alongside document text. Compatible clients and image-capable models can inspect the visual evidence, including charts and diagrams.
See what a connected AI can readConnect Claude and Claude Code
Fixed issues that prevented Claude Connect and Claude Code from completing their connection to a Mempipe Space. If a previous attempt failed, try connecting again.
Connect Claude or Claude CodeClearer search results for connected agents
MCP search results now keep the passage, source location, and citation link together in a structured response, giving agents a direct way to open the surrounding document text.
How agents searchFile processing recovers from interrupted saves
Fixed a problem where an interruption while saving processed document or table files could leave processing stuck on repeated retries.
If a file seems stuck