MCP BackEnd developer
We are looking for MCP BackEnd developer to join Unikie Team.
The Mission is to ensure the MCP servers are fed by clean, high-quality data and that the model's tool-use performance is constantly being monitored and tuned.
Responsibilities
- Design and maintain MCP Servers (Python/TypeScript) that expose internal data and functions to the model.
- Author rigorous JSON Schemas for tool definitions to ensure high model accuracy and reduce hallucinations.
- Architect secure Transport Layers (SSE/stdio) to bridge the gap between AI clients and remote data sources.
- Implement Least Privilege authorization logic to ensure the AI only accesses the data it absolutely needs.
Core Requirements: Expert-level knowledge of JSON-RPC, REST APIs, and Security protocols. Experience with LLM function-calling behaviors and prompt orchestration.
Expanded Responsibilities
- Semantic Data Engineering: Curate the datasets that power MCP tools. This involves cleaning unstructured data so that the Search or Fetch tools return high-signal results.
- Evaluation Pipelines (LLM-as-a-Judge): Build automated testing suites to score how well the LLM uses the MCP tools. If the model fails to call a tool correctly 10% of the time, you identify if the schema or the data is the bottleneck.
- Prompt Versioning & Registry: Manage a centralized repository of system prompts and tool definitions, ensuring that a change in the backend data schema doesn't break the model's ability to reason.
- Inference Monitoring: Track Time to First Token and Tool Execution Latency to optimize the costs and speed of the MCP backend.
Added Skills: LangSmith/Weights & Biases (for monitoring), Vector Database management (Pinecone/Weaviate/pgvector), and Python-based ETL pipelines.
The selected candidates will be invited to an interview for further discussion.
