MCP Server
Connect directly to the Pending AI platform from any AI Agent
Model Context Protocol
Pending AI offers a Model Context Protocol (MCP) server for accessing the different tool options. An MCP server gives existing LLM chats a way to call services enabled for a user and streamline different workflows. There are many possible workflow integrations, for example:
As Pending AI continues to deliver new possibilities, when in doubt you can always ask "What am I able to do using Pending AI?" for more information.
Supported Tools
The Pending AI MCP server currently supports the following tools in Early Access:
list_retrosynthesis_engines
List retrosynthesis engines. Use this tool to show available machine learning architectures and search strategies that can be used when submitting retrosynthesis jobs. Engine choice can impact the quality and diversity of retrosynthetic routes found.
list_retrosynthesis_libraries
List retrosynthesis libraries. Use this tool to show available building block libraries that can be used when submitting retrosynthesis jobs. Libraries specify collections of molecules that are available starting materials for a retrosynthetic search.
list_retrosynthesis_jobs
List retrosynthesis jobs. Use this tool to get short summaries of existing jobs, or to find a job_id based on SMILES and creation timestamps. Pagination is supported via offset-based page and limit parameters.
create_retrosynthesis_job
Create a retrosynthesis job. Use this tool to submit a molecule for retrosynthesis screening and provide optional parameters to customise the search behaviour. The molecule should be provided in SMILES format. A job_id is returned that can be used to poll the job status and retrieve results once it has completed.
retrieve_retrosynthesis_job
Retrieve a retrosynthesis job. Use this tool to fetch the details of a previously created retrosynthesis job by its id. Each job contains the molecule SMILES, status, and added timestamps for simple summary information.
retrieve_retrosynthesis_job_status
Retrieve the status of a retrosynthesis job. Expect a value that changes over time from submitted (waiting to be processed) to processing (actively being processed) to completed or failed (finished processing). Use this tool for polling the status of a job before before results become available.
retrieve_retrosynthesis_job_result
Retrieve the results of a previously submitted retrosynthesis job by its id. The list of retrosynthetic routes that were computed contain the synthetic steps to make the molecule with relevant building blocks. See the documentation for details about how retrosynthetic routes are represented.
delete_retrosynthesis_job
Delete a retrosynthesis job and generated results. The job will no longer be accessible. The operation is irreversible and should be used with caution. You cannot delete a job that is still processing.
list_molecule_generator_models
List generative models for sampling molecules. Use this tool to retrieve models and their metadata that can be used to see if a model is online and ready for sampling.
create_molecule_generator_sample
Sample molecules from a generative model. Molecules are represented in SMILES format and contain drug-like properties unique to the model used. Use this tool to generate novel structures that can be further filtered and evaluated. Conditional filtering should be applied after sampling to select relevant molecules.
Authentication
The MCP server aligns with current OAuth2.0 authentication flows allows for Dynamic Client Registration. Upon connecting to the server from the integrated tool connectors such as Claude or VSCode, you will be prompted to log in with your Pending AI account without any further steps.
If you are connecting from a service that is not one of the supported connectors, contact our team so that we can help provide any required callback URLs for your particular use case. For ease of testing, the MCP server already supports a callback on http://localhost:8000/auth/callback as used by most frameworks.
Last updated
Was this helpful?

