How japisuru/aucterra-mcp works
Aucterra MCP exposes document classification and structured field extraction to LLM agents over stdio. The workflow starts with an agent invoking a registered tool; each tool normalizes a local-file, URL, or base64 input, then sends a request to the corresponding Aucterra API and returns structured JSON or an error. The main entry point starts the MCP server. Both tools and input preparation live in server.py, so their shared implementation path is shown. The API services are documented by endpoint constants in that source; no other internal stages are evident.
Main entry points: server.py Field extraction · main.py stdio entry point
Components 8
MCP interface
- stdio entry point
aucterra_mcp/main.py - MCP tool server
aucterra_mcp/server.py
Document handling
- Input preparation
aucterra_mcp/server.py - Field extraction
aucterra_mcp/server.py - Document classification
aucterra_mcp/server.py
Aucterra services
- Extraction API
- Classification API
Other components
- LLM agent
Connections 7
- LLM agent → stdio entry point: starts stdio
README.md - stdio entry point → MCP tool server: runs server
aucterra_mcp/main.py - LLM agent → MCP tool server: invokes tools
README.md - Field extraction → Input preparation: normalizes input
aucterra_mcp/server.py - Document classification → Input preparation: normalizes input
aucterra_mcp/server.py - Field extraction → Extraction API: posts payload
aucterra_mcp/server.py - Document classification → Classification API: posts payload
aucterra_mcp/server.py