flexorch-mcp is the MCP server for FlexOrch, positioned as a "machine-facing SDK." It is used to connect Claude and other MCP-compatible agents to FlexOrch's document intelligence pipeline. Through natural-language-style tool calls, it can process documents, extract structured data, detect PII, and export LLM-ready datasets.
In terms of purpose, this tool mainly addresses the connection problem between agents and document data: agents directly initiate document processing and data extraction tasks, without the need for humans to write code to call the API. The objects it processes include business documents, such as PDFs, invoices, contracts, and pay slips, and the goal is to turn them into structured datasets, including steps such as classification, field extraction, PII redaction, and quality scoring.
In terms of main capabilities, this package is a lightweight proxy that exposes the FlexOrch API as MCP tools. The actual processing all takes place on FlexOrch's managed infrastructure. It supports document processing, structured data extraction, PII detection, and export of LLM-ready datasets. The API key is read from the environment variable FLEXORCH_API_KEY and is never passed as a tool parameter. The server is stateless and does not store or cache any data. PII redaction is completed by the FlexOrch pipeline before results are returned. All communication with api.flexorch.com uses HTTPS. All FlexOrch plan limits also apply to MCP tool calls, consuming quota per document processed.
In terms of usage, a FlexOrch account and API key are required, and the API key can be obtained from app.flexorch.com/settings. The runtime requires Python 3.10 or later. It can be added to the Claude Desktop configuration file (on macOS the path is ~/Library/Application Support/Claude/claude_desktop_config.json, on Windows the path is %APPDATA%\Claude\claude_desktop_config.json, and you can create the file yourself if it does not exist); it can also be added to Cursor's MCP configuration, or to ~/.codex/config.toml. After the connection configuration is complete, connection verification can be performed. The material also provides asample agent workflow.
In terms of applicable scenarios, it is suitable for using Claude or other MCP-compatible agents for batch document processing, structured information extraction, PII detection, and data export, and is especially suitable for agent workflows that need to operate in natural language rather than directly writing code to call it. For developers who need to write code, the material recommends using flexorch-sdk (Python) or flexorch-sdk-js (TypeScript) instead. There is also flexorch-audit, which can independently perform PII detection and document quality scoring without an account; complete documentation is available at docs.flexorch.com.
In terms of limitations, using this service requires a FlexOrch account and API key; all processing depends on FlexOrch's managed infrastructure; plan limits apply, consuming quota per document; the package itself does not store or cache data. Regarding specific feature details, charging, and pricing facts, the material does not provide them. The license is MIT.
