Overview
The Model Context Protocol (MCP) allows Benchling to connect securely with large language models (LLMs) and AI assistants, enabling scientists to query their Benchling data conversationally and receive traceable, structured results.
The Benchling MCP extends the capabilities of Benchling AI and Deep Research, allowing teams to access their R&D data within external AI tools while maintaining the same security, permissions, and audit controls as in Benchling.
What is the Benchling MCP?
The Benchling MCP is a standardized connection framework that lets AI chat applications securely access Benchling data.
Through the MCP, users can:
- Ask questions about their scientific data using natural language.
- Retrieve structured insights such as summaries, tables, or comparisons.
- Click back into Benchling for full traceability of sources and results.
Benchling MCP ensures all data access happens under the same governance, authentication, and permission rules that already exist in Benchling. No data is stored externally or used to train AI models.
How It Works
When a user submits a query through an LLM connected to Benchling (for example, “Summarize my last in vivo study”), the MCP:
- Uses the Benchling's agentic APIs to retrieve relevant data from Benchling.
- Returns an answer or report with information from Benchling to the LLM.
The LLM can then use that information retrieved from Benchling to continue executing its task
Prerequisites
To access Benchling through an MCP-compatible client, the following must be enabled in your Benchling tenant:
- Benchling Deep Research
- API access (V3 APIs - Alpha and Beta endpoints)
- API key for authentication
- Benchling MCP
If you don't have access to any of these, please contact your tenant admin and/or your Benchling point of contact.
Using the Benchling MCP
The Benchling MCP can be used by any AI agent or large language model that supports the Model Context Protocol (MCP). This makes it possible for Benchling Enterprise users and developers to build their own internal copilots, chat agents, or applications on top of Benchling data while maintaining security and governance.
This allows teams to create custom AI agents that can:
- Ask domain-specific scientific questions in natural language.
- Pull structured data from Benchling (e.g., results, notebook entries, or inventory items).
- Combine Benchling data with other systems, such as data lakes and other tools in your tech stack.
Example Scenarios
- Internal Copilot: Connect your in-house AI assistant to Benchling via the MCP to surface data summaries, compare experiments, or draft reports.
- Custom AI Interface: Build a model-driven analysis portal that uses Benchling as its secure data source.
- Third-party Model Integration: Integrate Benchling data into another MCP-based AI platform under your organization’s control.
Setup Overview
- Ensure you meet the pre-requisites listed above
- Follow the instructions of your MCP client
- Enter your MCP URL, which will follow the pattern: https://<tenant>.mcp.benchling.com/2025-06-18/mcp
- The MCP client should redirect you to sign into your Benchling tenant
- Ensure your tenant has Benchling AI (Deep Research) and V3 APIs enabled.
- Query Benchling securely through the MCP connection.
Availability in other MCP Clients
Claude Pro or Enterprise users can connect directly to Benchling through the Benchling MCP to securely query their scientific data.
More information on a 1-click setup in Claude coming soon.
Troubleshooting
If users see a 404 error when attempting to query Benchling through the MCP (for example, “Benchling MCP server isn’t properly connected”), this typically indicates that one or more required features are not enabled in their Benchling tenant.
Common causes:
- Benchling AI (Deep Research) is not enabled.
- V3 alpha APIs are not enabled for the tenant.
- The Benchling MCP is not properly configured in the corresponding client.
- The user’s API key or domain was entered incorrectly.
Permissions and Security
MCP queries run on behalf of the authenticated user, inheriting Benchling permissions and audit trails. Data remains within the customer’s Benchling tenant at all times. Claude and other AI providers do not use Benchling data for model training. Some responses include direct Benchling links for traceability.
For more details, see Security and Privacy for AI at Benchling.
Version History
- Previous versions of this article described the setup of a local MCP (also referred to as a "Desktop Extension"). We do not recommend using the local MCP version and strongly recommend upgrading to the remote MCP using the steps listed above.