M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
D

Docling MCP

Official
Making Docling agentic through MCP
How to install ↓ Compare GitHub source repository ↗
Category
Other #150 of 230
Stars
★ 769 Very popular
Transport
stdio (local process) · SSE · Streamable HTTP
Runtime
Python
Credentials
Optional API key
License
MIT
Last commit
Tools
18
50FMRS · D

Docling MCP is a feature-rich document conversion server supporting remote and local modes, offering a wide range of tools (conversion, generation, RAG). It suits developers integrating document processing into AI agents, but configuration can be complex and remote mode depends on an external service.

Strongest · Setup experience 12/20 Weakest · Reliability 6/20

Reliability
6/20
Security and permissions
10/20
Maintenance
10/20
Documentation
12/20
Setup experience
12/20
Why each score
Reliability 6/20
The manifest declares a coherent PyPI package, runtime arguments, and stdio transport, and the README describes conversion/generation tools and launch commands. However, no source code, CI workflows, or committed tests were supplied, so the MCP init/tool-list handshake and actual tool behavior cannot be verified. Also, the default conversion mode is remote while the required service URL is not marked required in the manifest, making a zero-config startup likely to fail. Score is lowered for missing execution evidence and error-path coverage.
Security and permissions 10/20
No malware, credential theft, destructive defaults, or real tokens/secrets appear in the supplied material; the API key is explicitly flagged as secret and provided via environment variables. Remote mode and external network use are disclosed in the README. However, only the manifest and README were reviewed, not the source code, so least-privilege implementation, data-flow boundaries, and confirmation of network operations cannot be confirmed. No red-line issue was triggered.
Maintenance 10/20
The repository is not archived, is MIT licensed, and the README shows a compatibility matrix and migration guide, indicating ongoing maintenance and versioning. But the supplied evidence lacks commit history, issue response metrics, dependency update policy, and a security-response channel. Stars and badges are discovery signals only and do not add points, so the score reflects missing governance evidence.
Documentation 12/20
The README is well structured with install modes, environment variable tables, client config JSON examples, and usage prompts. It lacks a troubleshooting section, explicit limits/cost disclosure, and a complete tool parameter reference. There is also an inconsistency: remote is the default mode and needs a service URL, but the manifest marks that variable as optional. These gaps and contradictions prevent a higher score.
Setup experience 12/20
Setup is presented as a short pip/uvx command with several ready-to-paste client configs, and remote, local, and hybrid modes are explained clearly. It remains a manual setup: remote mode requires an external Docling Serve instance, the default remote mode fails without a service URL, and no CI or committed test evidence verifies the steps. Under the static calibration cap, a mid-range score is appropriate.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesWrites / deletes local filesUses the networkConnects to a database

Best for

  • Developers looking to integrate document conversion into AI agents or MCP clients like Claude Desktop.
  • Teams needing PDF-to-Markdown/JSON conversion for RAG or knowledge management.
  • Users with access to Docling Serve or wanting hybrid mode with local fallback.

Not for

  • Pure document conversion users who do not need MCP integration (can use Docling directly).
  • Scenarios without any AI agent or MCP client.
  • Lightweight single-shot conversion with no caching requirements.

Required permissions

  • Read local files (access PDFs for conversion).
  • Network access (required for remote mode; optional for local mode).
  • Environment variables (must set DOCLING_MCP_* variables).
  • Storage permissions (for saving generated documents or caching).

Risks and side effects

  • Remote mode depends on Docling Serve; unavailability or high latency can cause failures.
  • API keys may be abused if leaked; handle with care.
  • Local conversion mode relies on system resources; large documents may cause high memory usage.
  • OCR and table structure detection may produce inaccurate results; verify outputs.

Setup

Before you start

Runtime：Python

DOCLING_MCP_SERVICE_API_KEY optionalsecret API key for authenticating with Docling Serve, provided by your Docling Serve deployment.
DOCLING_MCP_LI_API_KEY optionalsecret API key for that LLM endpoint; default none.
Other optional settings (15)
DOCLING_MCP_CONVERSION_MODE optional Conversion mode: 'remote' (default) or 'local' (requires docling-mcp[local] extra).
DOCLING_MCP_SERVICE_URL optional Base URL of your Docling Serve instance; required in remote mode, from docling-serve deployment or a managed SaaS.
DOCLING_MCP_SERVICE_TIMEOUT optional Request timeout in seconds for the remote Docling Serve API; default 300.0.
DOCLING_MCP_SERVICE_MAX_RETRIES optional Maximum retry attempts for remote API requests; default 3.
DOCLING_MCP_FALLBACK_TO_LOCAL optional Set to 'true' to fall back to local conversion when Docling Serve is unreachable (requires docling-mcp[local]).
DOCLING_MCP_KEEP_IMAGES optional Set to 'true' to retain page images in converted documents; default false.
DOCLING_MCP_IMAGES_SCALE optional Image scale factor (e.g. 1.0, 2.0); increase to avoid tensor batching errors; default 1.0.
DOCLING_MCP_DO_OCR optional Set to 'false' to disable the OCR pipeline; default true.
DOCLING_MCP_DO_TABLE_STRUCTURE optional Set to 'false' to disable table structure detection; default true.
DOCLING_MCP_LI_API_BASE optional OpenAI-compatible LLM endpoint for the llama-index-rag toolgroup; default http://127.0.0.1:1234/v1.
DOCLING_MCP_LI_MODEL_ID optional LLM model identifier; default ibm/granite-3.2-8b.
DOCLING_MCP_LI_EMBEDDING_MODEL optional HuggingFace embedding model; default BAAI/bge-base-en-v1.5.
DOCLING_MCP_LLS_URL optional LlamaStack server URL; default http://localhost:8321.
DOCLING_MCP_LLS_VDB_EMBEDDING optional Embedding model for the vector DB; default all-MiniLM-L6-v2.
DOCLING_MCP_LLS_EXTRACTION_MODEL optional Model used for structured extraction; default openai/gpt-oss-20b.
  1. Install the package: pip install docling-mcp or pip install docling-mcp[local].
  2. Configure environment variables as needed (e.g., DOCLING_MCP_CONVERSION_MODE, DOCLING_MCP_SERVICE_URL, etc.).
  3. Start the server: uvx --from docling-mcp docling-mcp-server --transport stdio.
claude_desktop_config.json 
{
  "mcpServers": {
    "docling": {
      "command": "uvx",
      "args": [
        "--from=docling-mcp",
        "docling-mcp-server"
      ],
      "env": {
        "DOCLING_MCP_CONVERSION_MODE": "remote",
        "DOCLING_MCP_SERVICE_URL": "https://your-docling-service.example.com",
        "DOCLING_MCP_SERVICE_API_KEY": "your-api-key-here"
      }
    }
  }
}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "docling": {
      "command": "uvx",
      "args": [
        "--from=docling-mcp",
        "docling-mcp-server"
      ],
      "env": {
        "DOCLING_MCP_CONVERSION_MODE": "remote",
        "DOCLING_MCP_SERVICE_URL": "https://your-docling-service.example.com",
        "DOCLING_MCP_SERVICE_API_KEY": "your-api-key-here"
      }
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add docling -e DOCLING_MCP_CONVERSION_MODE=remote -e DOCLING_MCP_SERVICE_URL=https://your-docling-service.example.com -e DOCLING_MCP_SERVICE_API_KEY=your-api-key-here -- uvx --from=docling-mcp docling-mcp-server

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

After launch, check that tools like convert_pdf_to_docling_document appear in the client's tool list, then ask it to convert a PDF into a DoclingDocument and return its document-key to confirm the connection works.

Troubleshooting

  1. Check that `DOCLING_MCP_CONVERSION_MODE` is 'remote' and `DOCLING_MCP_SERVICE_URL` is correct.
  2. Ensure the API key is valid and has permissions.
  3. For remote mode, verify network connectivity and firewall settings.
  4. If encountering tensor batching errors, try increasing `DOCLING_MCP_IMAGES_SCALE`.
  5. Check server logs for detailed error messages.

Things to try

Once connected, you can ask your AI assistant things like:

  • Convert the PDF document at <provide file-path> into DoclingDocument and return its document-key.
  • Convert this PDF URL to Markdown.
  • Create a new Docling document and write a report on the impact of tokenizers on LLM quality, with a title, headings, paragraphs and a list, then save it and return the filepath.
  • Export the current Docling document to HTML.

Tools 18

convert_pdf_to_docling_document read-only
Convert a local PDF document to structured DoclingDocument format.
convert_pdf_to_docling_document_from_url read-only
Convert a PDF from a remote URL to DoclingDocument format.
create_new_docling_document writes
Create a new empty DoclingDocument.
add_title_to_docling_document writes
Add a title to the document.
add_heading_to_docling_document writes
Add a section heading to the document.
add_paragraph_to_docling_document writes
Add a paragraph to the document.
open_list_in_docling_document writes
Open a list within the document.
add_listitem_to_list_in_docling_document writes
Add a list item to the currently opened list.
Show 10 more tools
close_list_in_docling_document writes
Close the currently opened list.
add_table_to_docling_document writes
Add a table to the document.
export_docling_document_to_markdown read-only
Export the document to Markdown format.
export_docling_document_to_html read-only
Export the document to HTML format.
save_docling_document writes
Save the document to a file.
retrieve_docling_document read-only
Retrieve a document by its document key from cache.
convert_to_markdown read-only
Convert a PDF directly to Markdown format.
milvus_upload writes
Upload a document to Milvus vector database.
milvus_search read-only
Perform retrieval in Milvus.
llama_index_rag_search read-only
Perform RAG retrieval using LlamaIndex.

Use cases

Convert PDF documents to structured JSON for AI applications.
Create, edit, and export Docling documents via AI agents.
Embed documents into vector databases (e.g., Milvus) for RAG pipelines.
Provide local or remote document conversion for LLMs.

Supported clients

Claude Desktop
LM Studio
Llama Stack

Listed from the project's documentation, not tested by this site.

Overview

Docling MCP is a document processing service using the Model Context Protocol (MCP) for tool integration. It leverages the Docling library to convert PDF and other documents into structured formats, providing conversion, generation, and RAG tools. Supports remote (Docling Serve) and local (docling-mcp[local]) conversion modes.

Similar servers

BioMCP 75 · B

One binary. One grammar. Evidence from the biomedical sources you already trust.

★ 654 · Tools 11 Compare with this →
Compare with other MCP servers

Source revision 5e1551a9b984 Data synced 2026-10-12 Read the FMRS scoring method