Unstructured MCP Server

What is Unstructured MCP Server?

A server that exposes operations for interacting with the Unstructured API. It lets you manage data sources (e.g., S3, Azure, Google Drive), destinations (e.g., Weaviate, Pinecone, MongoDB), define workflows that move data between them, and run or monitor jobs.

How to use Unstructured MCP Server?

1. Install the Python package:

   uv pip install uns_mcp
   

2. Configure your environment variables (e.g., UNSTRUCTURED_API_KEY, connector credentials) in a .env file or your shell.
3. Start the server using one of the supported commands, for example:

   uvx uns_mcp            # recommended short‑hand
   # or
   python -m uns_mcp      # editable‑package style
   

4. Connect a client (Claude Desktop, minimal client, or any MCP‑compatible tool) to the server endpoint (http://127.0.0.1:8080/sse for SSE or the stdio protocol).
5. Invoke the desired tool names (list_sources, create_workflow, run_workflow, etc.) through the client.

Key Features

• Connector Management: List, create, update, and delete both source and destination connectors for a wide range of services (S3, Azure Blob, Google Drive, Salesforce, Neo4j, MongoDB, etc.).
• Workflow Orchestration: Define workflows that bind a source to a destination, run them on demand, and retrieve detailed configuration.
• Job Control: List jobs per workflow, get job status, and cancel running jobs where supported.
• Firecrawl Integration: Built‑in web‑crawling and LLM‑optimized text extraction via Firecrawl.
• Environment‑Driven Credentials: All connector secrets are supplied through .env variables, keeping sensitive data out of code.
• Multiple Communication Protocols: Supports SSE, stdio, and direct CLI invocation, making it flexible for various client ecosystems.

Use Cases

• Data Ingestion Pipelines: Pull documents from S3 or SharePoint, transform them, and store vectors in Pinecone or Weaviate.
• Periodic Syncs: Schedule workflows that keep a MongoDB collection in sync with a Google Drive folder.
• Web Content Harvesting: Use Firecrawl to crawl a website, generate LLM‑ready text, and dump the output to an S3 bucket for further processing.
• Enterprise Integration: Connect Salesforce objects to a Neo4j graph for relationship analysis.
• Rapid Prototyping: Developers can spin up a local server and experiment with connector configurations before moving to production.

FAQ

Q: Which Python version is required? A: Python 3.12 or newer.

Q: Do I need the uv tool? A: It is the recommended environment manager, but you can use any virtual‑env tool (venv, poetry, conda) as long as the dependencies are installed.

Q: How are credentials handled? A: Set them as environment variables (e.g., AWS_KEY, AWS_SECRET, WEAVIATE_CLOUD_API_KEY) or place them in a .env file referenced by the server.

Q: Can I run the server without Claude Desktop? A: Yes. Use the minimal client (minimal_client/client.py) or any custom MCP client.

Q: What if I need to debug API calls? A: Set DEBUG_API_REQUESTS=true to log request parameters to unstructured-client-<date>.log.

Q: Is there support for cancelling Firecrawl LLM‑text jobs? A: Currently the Firecrawl API does not allow cancellation of LLM‑text jobs; the provided cancel_llmtxt_job is a placeholder.

Unstructured API MCP Server

An MCP server implementation for interacting with the Unstructured API. This server provides tools to list sources and workflows.

Available Tools

Tool	Description
list_sources	Lists available sources from the Unstructured API.
get_source_info	Get detailed information about a specific source connector.
create_source_connector	Create a source connector.)
update_source_connector	Update an existing source connector by params.
delete_source_connector	Delete a source connector by source id.
list_destinations	Lists available destinations from the Unstructured API.
get_destination_info	Get detailed info about a specific destination connector
create_destination_connector	Create a destination connector by params.
update_destination_connector	Update an existing destination connector by destination id.
delete_destination_connector	Delete a destination connector by destination id.
list_workflows	Lists workflows from the Unstructured API.
get_workflow_info	Get detailed information about a specific workflow.
create_workflow	Create a new workflow with source, destination id, etc.
run_workflow	Run a specific workflow with workflow id
update_workflow	Update an existing workflow by params.
delete_workflow	Delete a specific workflow by id.
list_jobs	Lists jobs for a specific workflow from the Unstructured API.
get_job_info	Get detailed information about a specific job by job id.
cancel_job	Delete a specific job by id.
list_workflows_with_finished_jobs	Lists all workflows that have any completed job, together with information about source and destination details.

Below is a list of connectors the UNS-MCP server currently supports, please see the full list of source connectors that Unstructured platform supports here and destination list here. We are planning on adding more!

Source	Destination
S3	S3
Azure	Weaviate
Google Drive	Pinecone
OneDrive	AstraDB
Salesforce	MongoDB
Sharepoint	Neo4j
	Databricks Volumes
	Databricks Volumes Delta Table

To use the tool that creates/updates/deletes a connector, the credentials for that specific connector must be defined in your .env file. Below is the list of credentials for the connectors we support:

Credential Name	Description
ANTHROPIC_API_KEY	required to run the minimal_client to interact with our server.
AWS_KEY, AWS_SECRET	required to create S3 connector via uns-mcp server, see how in documentation and here
WEAVIATE_CLOUD_API_KEY	required to create Weaviate vector db connector, see how in documentation
FIRECRAWL_API_KEY	required to use Firecrawl tools in external/firecrawl.py, sign up on Firecrawl and get an API key.
ASTRA_DB_APPLICATION_TOKEN, ASTRA_DB_API_ENDPOINT	required to create Astradb connector via uns-mcp server, see how in documentation
AZURE_CONNECTION_STRING	required option 1 to create Azure connector via uns-mcp server, see how in documentation
AZURE_ACCOUNT_NAME+AZURE_ACCOUNT_KEY	required option 2 to create Azure connector via uns-mcp server, see how in documentation
AZURE_ACCOUNT_NAME+AZURE_SAS_TOKEN	required option 3 to create Azure connector via uns-mcp server, see how in documentation
NEO4J_PASSWORD	required to create Neo4j connector via uns-mcp server, see how in documentation
MONGO_DB_CONNECTION_STRING	required to create Mongodb connector via uns-mcp server, see how in documentation
GOOGLEDRIVE_SERVICE_ACCOUNT_KEY	a string value. The original server account key (follow documentation) is in json file, run base64 < /path/to/google_service_account_key.json in terminal to get the string value
DATABRICKS_CLIENT_ID,DATABRICKS_CLIENT_SECRET	required to create Databricks volume/delta table connector via uns-mcp server, see how in documentation and here
ONEDRIVE_CLIENT_ID, ONEDRIVE_CLIENT_CRED,ONEDRIVE_TENANT_ID	required to create One Drive connector via uns-mcp server, see how in documentation
PINECONE_API_KEY	required to create Pinecone vector DB connector via uns-mcp server, see how in documentation
SALESFORCE_CONSUMER_KEY,SALESFORCE_PRIVATE_KEY	required to create salesforce source connector via uns-mcp server, see how in documentation
SHAREPOINT_CLIENT_ID, SHAREPOINT_CLIENT_CRED,SHAREPOINT_TENANT_ID	required to create One Drive connector via uns-mcp server, see how in documentation
LOG_LEVEL	Used to set logging level for our minimal_client, e.g. set to ERROR to get everything
CONFIRM_TOOL_USE	set to true so that minimal_client can confirm execution before each tool call
DEBUG_API_REQUESTS	set to true so that uns_mcp/server.py can output request parameters for better debugging

Firecrawl Source

Firecrawl is a web crawling API that provides two main capabilities in our MCP:

1. HTML Content Retrieval: Using invoke_firecrawl_crawlhtml to start crawl jobs and check_crawlhtml_status to monitor them
2. LLM-Optimized Text Generation: Using invoke_firecrawl_llmtxt to generate text and check_llmtxt_status to retrieve results

How Firecrawl works:

Web Crawling Process:

• Starts with a specified URL and analyzes it to identify links
• Uses the sitemap if available; otherwise follows links found on the website
• Recursively traverses each link to discover all subpages
• Gathers content from every visited page, handling JavaScript rendering and rate limits
• Jobs can be cancelled with cancel_crawlhtml_job if needed
• Use this if you require all the info extracted into raw HTML, Unstructured's workflow cleans it up really well :smile:

LLM Text Generation:

• After crawling, extracts clean, meaningful text content from the crawled pages
• Generates optimized text formats specifically formatted for large language models
• Results are automatically uploaded to the specified S3 location
• Note: LLM text generation jobs cannot be cancelled once started. The cancel_llmtxt_job function is provided for consistency but is not currently supported by the Firecrawl API.

Note: A FIRECRAWL_API_KEY environment variable must be set to use these functions.

Installation & Configuration

This guide provides step-by-step instructions to set up and configure the UNS_MCP server using Python 3.12 and the uv tool.

Prerequisites

• Python 3.12+
• uv for environment management
• An API key from Unstructured. You can sign up and obtain your API key here.

Using uv (Recommended)

No additional installation is required when using uvx as it handles execution. However, if you prefer to install the package directly:

uv pip install uns_mcp

Configure Claude Desktop

For integration with Claude Desktop, add the following content to your claude_desktop_config.json:

Note: The file is located in the ~/Library/Application Support/Claude/ directory.

Using uvx Command:

{
   "mcpServers": {
      "UNS_MCP": {
         "command": "uvx",
         "args": ["uns_mcp"],
         "env": {
           "UNSTRUCTURED_API_KEY": "<your-key>"
         }
      }
   }
}

Alternatively, Using Python Package:

{
   "mcpServers": {
      "UNS_MCP": {
         "command": "python",
         "args": ["-m", "uns_mcp"],
         "env": {
           "UNSTRUCTURED_API_KEY": "<your-key>"
         }
      }
   }
}

Using Source Code

1. Clone the repository.

2. Install dependencies:

   uv sync
   

3. Set your Unstructured API key as an environment variable. Create a .env file in the root directory with the following content:

   UNSTRUCTURED_API_KEY="YOUR_KEY"
   

   Refer to .env.template for the configurable environment variables.

You can now run the server using one of the following methods:

uvx pip install -e .

Update your Claude Desktop config:

{
  "mcpServers": {
    "UNS_MCP": {
      "command": "uvx",
      "args": ["uns_mcp"]
    }
  }
}

Note: Remember to point to the uvx executable in environment where you installed the package

Note: Not supported by Claude Desktop.

For SSE protocol, you can debug more easily by decoupling the client and server:

1. Start the server in one terminal:

   uv run python uns_mcp/server.py --host 127.0.0.1 --port 8080
   # or
   make sse-server
   

2. Test the server using a local client in another terminal:

   uv run python minimal_client/client.py "http://127.0.0.1:8080/sse"
   # or
   make sse-client
   

Note: To stop the services, use Ctrl+C on the client first, then the server.

Configure Claude Desktop to use stdio:

{
  "mcpServers": {
    "UNS_MCP": {
      "command": "ABSOLUTE/PATH/TO/.local/bin/uv",
      "args": [
        "--directory",
        "ABSOLUTE/PATH/TO/YOUR-UNS-MCP-REPO/uns_mcp",
        "run",
        "server.py"
      ]
    }
  }
}

Alternatively, run the local client:

uv run python minimal_client/client.py uns_mcp/server.py

Additional Local Client Configuration

Configure the minimal client using environmental variables:

• LOG_LEVEL="ERROR": Set to suppress debug outputs from the LLM, displaying clear messages for users.
• CONFIRM_TOOL_USE='false': Disable tool use confirmation before execution. Use with caution, especially during development, as LLM may execute expensive workflows or delete data.

Debugging tools

Anthropic provides MCP Inspector tool to debug/test your MCP server. Run the following command to spin up a debugging UI. From there, you will be able to add environment variables (pointing to your local env) on the left pane. Include your personal API key there as env var. Go to tools, you can test out the capabilities you add to the MCP server.

mcp dev uns_mcp/server.py

If you need to log request call parameters to UnstructuredClient, set the environment variable DEBUG_API_REQUESTS=false. The logs are stored in a file with the format unstructured-client-{date}.log, which can be examined to debug request call parameters to UnstructuredClient functions.

Add terminal access to minimal client

We are going to use @wonderwhy-er/desktop-commander to add terminal access to the minimal client. It is built on the MCP Filesystem Server. Be careful, as the client (also LLM) now has access to private files.

Execute the following command to install the package:

npx @wonderwhy-er/desktop-commander setup

Then start client with extra parameter:

uv run python minimal_client/client.py "http://127.0.0.1:8080/sse" "@wonderwhy-er/desktop-commander"
# or
make sse-client-terminal

Using subset of tools

If your client supports using only subset of tools here are the list of things you should be aware:

• update_workflow tool has to be loaded in the context together with create_workflow tool, because it contains detailed description on how to create and configure custom node.

Known issues

• update_workflow - needs to have in context the configuration of the workflow it is updating either by providing it by the user or by calling get_workflow_info tool, as this tool doesn't work as patch applier, it fully replaces the workflow config.

CHANGELOG.md

Any new developed features/fixes/enhancements will be added to CHANGELOG.md. 0.x.x-dev pre-release format is preferred before we bump to a stable version.

Troubleshooting

• If you encounter issues with Error: spawn <command> ENOENT it means <command> is not installed or visible in your PATH:
  • Make sure to install it and add it to your PATH.
  • or provide absolute path to the command in the command field of your config. So for example replace python with /opt/miniconda3/bin/python