Developers
Give agents grounded document tools
Claude, ChatGPT and Cursor call parsing as tools — and get citations back. Agent2Agent ready for orchestrator fleets.
Tools, not chat wrappers
01
Tool per endpoint
extract, parse, split, classify, batch status — one MCP tool each.
02
Citations included
Tool results carry page + bbox, so answers quote the source line.
03
Fleet ready
Agent2Agent protocol lets orchestrators call parsing agents directly.
What the server exposes
Pick a capability to see what it does in Ambriel.
total · 1.240,500.97
p.1 · bboxcited
extract_structured
File + schema in, cited JSON out — the whole API as one tool.
- Schema passthrough
- Evidence in results
- Sync + async
Connect an agent in minutes
01
Add the server
One JSON block in Claude/Cursor MCP config with your API key.
02
List the tools
Agent discovers extract, parse and batch tools automatically.
03
Ask with citations
“Total on INV-041?” returns 1.240,50 with page and box.
Related use cases
Any MCP client: Claude, ChatGPT (connectors), Cursor and Agent2Agent orchestrators. Config is one JSON block.
No — tool calls meter like API calls (4/page, +2 tables, −50% batch) with X-Credits-Used in results.
Yes — file bytes, file_url or a prior pipeline job id (jobid:// reuse like Reducto).
Phase 3: orchestrators call parsing agents directly with the same evidence contract. MCP today, A2A next.
Ground your agents today
Document tools with citations.
“Total on INV-041?” Claude calls the tool and answers 1.240,50 with the source box.