VectorShift MCP. Control your entire AI data workflow from one chat.

Listen up. VectorShift runs complex AI pipelines and knowledge retrieval (RAG) directly through your agent client. You don't have to switch screens or use a dozen different tools; you just talk to your agent, and it makes the whole data workflow happen.

Executing Multi-Step Pipelines

The core of what we do here is automation. You start by creating a multi-step automated process using create_pipeline, which registers a brand new workflow in the system. Once you've built that pipeline, you can run it with specific inputs using run_pipeline. Need to fire off the same job ten times at once? Just hit up bulk_run_pipeline and let it handle all those instances simultaneously.

You also manage the execution state: if a process gets stuck or needs a break, use pause_pipeline to stop it; when you're ready, kick it back on with resume_pipeline. If something goes sideways, you can kill any running instance forcefully using terminate_pipeline. To wrap things up and clean house, you can delete the whole workflow with delete_pipeline, or just get a full list of every pipeline definition with list_pipelines.

Querying Proprietary Knowledge Bases (RAG)

This is how you ground your AI in real company data. You first set up a secure knowledge container by calling create_knowledge_base. Once that's live, you feed it external data—whether those are files or URLs—using index_knowledge_base. The system then makes everything searchable. When the user asks a question, you don't run a keyword search; you perform a semantic search right against all the indexed documents using query_knowledge_base.

You can check what's inside any given container by running list_knowledge_base_documents, and you always know the status of your data set with get_knowledge_base or list_knowledge_bases. Don't forget, if a knowledge base is obsolete, you can permanently delete it using delete_knowledge_base. If you only need to prune some documents, you can remove specific files by their unique ID through delete_knowledge_base_documents.

Managing Dedicated Chatbot Sessions

For focused conversation, you use dedicated chatbots. You start by registering one with create_chatbot, and the system gives you a unique instance to work with. To talk to it, you send a message using run_chatbot. Need to give the chat more context for that specific session? Use upload_chatbot_files to attach relevant documents directly to the conversation thread.

You can get all your active bot details by running list_chatbots, and check on any single one with get_chatbot. If a chatbot is doing too much or you just need it shut down, you can terminate the session instantly using terminate_chatbot. To clean up, you can delete an entire instance with delete_chatbot.

Building Custom Logic and Transformations

Sometimes, talking to the AI isn't enough; you gotta manipulate the data itself. You build custom logic—using Python or JavaScript code—and register it by calling create_transformation. This lets your agent run complex, coded steps as part of a larger workflow. To see what code you've written up, use list_transformations, and check the configuration for any specific one with get_transformation.

Once you're ready to execute that custom data manipulation, hit run_transformation with your variables. If this logic turns out to be bad or unused, you can scrub it permanently using delete_transformation.

How It All Connects Together

You also have tools for the heavy lifting of monitoring and list-checking everything in the system: You can get configuration details on any pipeline with get_pipeline, or check out all available transformations via list_transformations. This gives you total visibility into every piece of automation, knowledge, and chatbot running under your account.