MCP Support
We are excited to announce that Langdock now supports the Model Context Protocol (MCP)! 🚀

What is MCP?
MCP is an open protocol that standardizes how AI applications connect to data sources and tools. Think of MCP like a USB-C port for AI applications. Just as USB-C provides a standardized way to connect your devices to various systems, MCP provides a standardized way to connect AI models to different data sources and tools.
More specifically, you can connect your assistants to remote MCP servers over the internet. This offers more integration possibilities with different data sources in addition to existing integrations.
Key Features
- Flexible Authentication: Works with both simple API key authentication and secure OAuth connections for services that require it.
- Automatic Tool Discovery: MCP connections automatically find what tools are available from each server.
- Seamless Integration: MCP tools work naturally within your assistant conversations (can require confirmation and display inputs and outputs).
- Continuous Permission Management: MCP integrations work within our existing integration framework, so you can manage permissions, sharing, and usage just like your other Langdock integrations.
- Trusted Compatibility: We verified compatibility with popular MCP servers including Microsoft Docs, Cloudflare Docs, and GitHub. Generally, we support all open MCP servers (without authentication) and MCP servers with API-Authentication.
How to Get Started with MCP
You can set up MCP integrations, through the "Add integration" button in the integration menu. Then, you can select if you want to build an integration with JavaScript (as before) or connect a remote MCP server. To set up MCP integrations, users need permission to create custom integrations.
Detailed instructions can be found on our docs:
https://docs.langdock.com/resources/integrations/mcp
Note: MCP is still in a very early stage of development. We strongly recommend thoroughly testing any integrations you build using MCP before relying on them in production.
- Answer generation continues in the background: The response generation continues when users navigate away from chat.
- Optimize assistant instructions with AI: You can now optimize your assistant instruction with AI or dictate your prompt by clicking on the buttons in the instruction field.
- Optimized Assistant configuration: We switched the places of the input type and instruction fields in the Assistant builder to improve UX
- Dismissible notifications in the top right corner: You can now click notifications in the bottom right corner away.
- Improved CSV and Excel content preview: We improved the preview and handling of spreadsheet files when uploading them to assistants and chats.
- Enhanced model description cards: The model selector now includes precise capabilities and short descriptions, making it easier to select models with reasoning, web search, and data analysis capabilities.
- Improved assistant form builder: Assistant forms now support up to 15 fields.
- Custom model test button: Custom models now can be tested in the workspace settings with a test button.
- Attachments without text prompts: You can now submit files directly without a prompt.
- Context Actions: Previously, you had to manually add dependent actions like fetching a channel ID before sending a Slack Message to a Group Chat so the assistant could perform this task correctly. This often meant trial and error to figure out why the assistant wasn’t working as expected. Now, for actions that require additional information like IDs, the action by default includes the necessary dependent actions. You will see this denoted by a small link icon on the action tile and when you open the action by clicking the small toggle.