Flexorch MCP

What flexorch/flexorch-mcp MCP server does

The flexorch/flexorch-mcp MCP server exposes the FlexOrch API as MCP tools for document intelligence workflows. It is intended for agents rather than application code: developers building directly against FlexOrch can use the separate Python or TypeScript SDKs referenced by the project.

The pipeline accepts PDF, DOCX, TXT, XLSX, HTML, XML, EML, JPG, PNG, and TIFF files. Processing can classify documents, extract structured fields, mask personally identifiable information across more than 10 locales, and provide quality scoring. The resulting data can be assembled into datasets for downstream model or retrieval workflows.

How it works

Processing is asynchronous. An agent first submits a document with document.process, then polls the resulting job using job.status. Once processing completes, job.result returns the extracted fields. document.reprocess can send an already-uploaded document through the pipeline again.

A completed execution can be passed to dataset.build to create a structured dataset. Datasets can be exported in JSONL, CSV, JSON, XML, or Markdown formats. The export operation also supports RAG output containing chunks intended for LangChain or LlamaIndex workflows.

For plans that include semantic features, dataset.index starts vector indexing, dataset.search performs semantic search over indexed datasets, and dataset.chunks returns paginated RAG-ready chunks.

Setup and configuration

Install the package with Python 3.10 or newer:

pip install flexorch-mcp

The flexorch/flexorch-mcp MCP server requires a FlexOrch account and an API key. Set the key in the FLEXORCH_API_KEY environment variable; it is not supplied as a tool argument. The documented executable is flexorch-mcp.

Claude Desktop and Cursor configurations run the executable directly and provide the environment variable. An OpenAI Codex example uses uvx flexorch-mcp with the same variable. The repository provides a flexorch-mcp --check command to verify the key, API connection, plan, and registered tools.

Tools and capabilities

The available tools cover the full document-to-dataset flow:

• document.process uploads and processes supported files.
• document.reprocess re-queues an uploaded document.
• job.status checks whether an asynchronous job completed or failed.
• job.result retrieves structured fields from a completed job.
• dataset.build creates a dataset from a completed execution.
• dataset.export returns a dataset in several structured or RAG-oriented formats.
• dataset.index, dataset.search, and dataset.chunks provide vector indexing, semantic search, and chunk retrieval on eligible plans.

Limitations and notes

The server is a stateless proxy: it does not store or cache document data. Processing takes place on FlexOrch-managed infrastructure, and communication with api.flexorch.com uses HTTPS. PII masking is applied by FlexOrch before results are returned.

Usage follows the FlexOrch plan rules, with credits consumed per processed document. Trial and Starter plans do not include semantic search; Pro includes it, while Enterprise limits are custom. The flexorch/flexorch-mcp MCP server therefore requires a plan with the relevant feature for indexing, search, or chunk retrieval. The project is licensed under MIT.