Harper MCP Server

Harper MCP Server – Bridging AI Assistants and HarperDB

The Harper MCP Server solves a common pain point for developers building AI‑powered applications: accessing structured data in a fast, secure, and standardized way. By exposing HarperDB tables as resources through the Model Context Protocol, it allows Claude and other AI assistants to query, list, and retrieve data without the assistant needing direct database credentials or custom connectors. This abstraction keeps the AI workflow simple while preserving HarperDB’s powerful geo‑distributed architecture and fine‑grained security.

At its core, the server implements two essential MCP methods: and . The former returns a catalog of all tables and custom resources that the authenticated user may access, each represented by a clear URI. The latter fetches data for a specific resource or row, supporting optional query parameters for filtering and pagination (, ). The server adheres strictly to JSON‑RPC 2.0, ensuring that every request and response is predictable and easily parsed by AI agents.

Key capabilities include:

• Read‑only data exposure – safeguarding the underlying database from unintended writes while still providing full read access.
• Query‑parameter filtering – enabling AI agents to request only relevant slices of data, reducing payload size and improving performance.
• Pagination support – allowing large tables to be consumed incrementally, which is essential for conversational agents that need to display lists or search results.
• Standardized error handling – giving AI assistants consistent feedback for authentication failures, missing resources, or malformed requests.

Developers can leverage Harper MCP Server in a variety of real‑world scenarios. For example, an AI assistant could walk users through their order history by listing the table and then reading individual orders on demand. In a knowledge‑base application, the server could expose custom resources that aggregate data from multiple tables, letting the assistant answer complex queries without additional backend logic. Because HarperDB already handles sharding and replication, the server scales automatically as data grows or as users spread across regions.

Integration is straightforward: a client simply posts JSON‑RPC requests to the single endpoint, using HarperDB’s native authentication mechanisms (Basic Auth, JWT, or mTLS). The server then translates those calls into HarperDB queries, applies any filters or pagination, and returns the results in a uniform JSON format. This tight coupling means that AI workflows can treat HarperDB like any other MCP‑compatible data source, unifying diverse backends under a single protocol and simplifying the development of AI‑centric applications.