Model Context Protocol (MCP) Server
The Knowledgator MCP Server connects AI applications, coding assistants, and autonomous agent runtimes (such as Claude Desktop, Cursor, and custom agent workflows) directly to Knowledgator's inference, dataset generation, and model fine-tuning services using the standard Model Context Protocol (MCP).
Quickstart & Connection
You can connect to the Knowledgator MCP server in two ways:
- Hosted Remote Endpoint (Recommended): Connect instantly via remote MCP (
https://mcp.knowledgator.com/mcp) with zero local installation. - Self-Hosted / Local Server: Run on your local machine or private VPC via
uvx, Git, or Docker.
👉 View the complete setup instructions, client JSON configurations, and environment variables in the Setup & Connection Guide.
Available MCP Tools
All tools are grouped by functional domain, mirroring the Knowledgator REST API and SDK modules:
1. Inference Tools
Run deterministic, hallucination-free information extraction and classification directly within your agentic workflows:
comprehend_inference: Zero-shot and few-shot text classification with confidence scores and threshold filtering.extract_inference: Open-vocabulary named entity recognition (NER) and relation extraction.instruct_inference: Instruction-driven structured information extraction using custom schema templates.open_model_inference: Direct inference execution on standalone full Open Models and custom checkpoints.
2. Dataset Tools
Manage datasets, inspect schemas, stream downloads, and upload data programmatically:
upload_dataset: Upload validated in-memory records or JSON strings to create a new dataset.download_dataset: Download dataset records statelessly in memory via storage redirect.generate_dataset_plan: Generate a structured dataset plan (taxonomy, sample schema, collection steps) from a prompt.start_dataset_generation: Trigger an asynchronous synthetic dataset generation pipeline.list_datasets: Query and filter platform datasets with pagination.get_dataset_details: Retrieve dataset metadata, schema, and sample counts.
3. Fine-Tuning Tools
Control model training experiments and monitor training progress from your coding assistant:
start_training_job: Launch a model or adapter fine-tuning job with custom hyperparameters.get_training_status: Poll job status, current step, loss, and training milestones.list_trainings: List active, queued, and completed fine-tuning jobs.stop_training_job: Stop or cancel a running training job.
Resources
The server exposes an informational MCP resource:
config://app: Returns runtime server configuration (server name, debug mode, auth type, and active base URL).
Next Steps
- Follow the Setup & Connection Guide to configure Claude Desktop, Cursor, or your custom agent harness.
- Explore individual tool pages above for parameter schemas, example prompts, and sample JSON payloads.