TechPrufer
Get started
Pinecone
Pinecone logoPinecone · Feature trail

Agent-Friendly API Design

2 posts · Jan 2025 – Aug 2026

What this trail covers

Pinecone Assistant is now generally available, offering optimized interfaces with new chat and context APIs, custom instructions for tailoring assistant behavior, expanded input/output formats (JSON, .md, .docx), and region control (EU/US). The Context API provides structured context for agentic workflows, and the Chat API delivers grounded responses with citations. Custom instructions allow for fine-tuning assistant roles, and metadata filters enhance data management. The service is built on Pinecone's managed vector database, ensuring data security and privacy.

All Pinecone trails
  1. Designing Agent-Friendly APIsThis post introduces six design principles for creating agent-friendly APIs: 1. Errors are guidance (actionable error messages with clear fixes and doc links, machine-readable error codes). 2. Budget the reader's context (bounded responses, pagination, caps, high-signal fields first, search over list, verifiable success in-band). 3. Self-description beats documentation (describe/capabilities endpoints, complete OpenAPI specs with realistic examples, loud deprecation). 4. Safe at machine tempo (idempotency as a contract, pace-able rate limits, predictable outcomes, guardrails like dry-run/preview). It also emphasizes the importance of documentation and an organizational lesson. The post introduces two key metrics: Turns-to-First-Successful-Call (TTFSC) and Unattended Task-Success Rate, and discusses diagnostic metrics like actionable error percentage and token cost per task. It shares findings from experiments with Pinecone's own API and Claude Sonnet 5, highlighting the impact of clear error messages and SDK design on agent success.
  2. Use the Pinecone Plugin for Claude Code to develop AI Applications FasterThis post introduces the Pinecone Plugin for Claude Code, making Pinecone's vector database and managed RAG service accessible directly within the Anthropic Claude Code IDE. It enables natural language and explicit commands for managing Pinecone indexes, upserting vectors, querying assistants, and generating Pinecone-related code. The plugin also facilitates the creation and management of Pinecone Assistants for RAG, including document upload, synchronization, and obtaining cited answers, all without leaving the Claude Code environment. This significantly streamlines the AI application development workflow by integrating Pinecone's capabilities into the developer's inner loop.
  3. Add context to your agent with Pinecone Assistant MCP remote serverThis post introduces the Pinecone Assistant's remote MCP server, enabling agents to easily integrate with Pinecone Assistant's context capabilities via the Model Context Protocol (MCP). It details how to set up an Assistant, upload documents, and connect it to an agent using Langchain. The post also explains how to configure Claude Desktop to use the remote MCP server via a local proxy like `supergateway`. This extends the agent-friendly API design by making remote server integration a first-class citizen.
  4. Tool up: Pinecone’s first MCP servers are hereThis post introduces three new MCP servers: Pinecone Assistant MCP (remote), Pinecone Assistant MCP (local), and Pinecone Developer MCP (local). The Pinecone Developer MCP includes a `search-docs` tool that leverages Pinecone's documentation to improve AI coding assistant output. The post details the MCP protocol's role in standardizing LLM tool calling and provides a Node.js code example demonstrating how to connect to the Pinecone Developer MCP and use its tools with an LLM.
  5. Easily build knowledgeable chat and agent-based applications in minutes with Pinecone Assistant, now generally availableThis post announces the General Availability (GA) of Pinecone Assistant. Key technical contributions include the introduction of a new Chat API for structured, grounded responses with citations (supporting streaming and batch modes) and a Context API that provides structured context (expanded chunks with relevancy scores and references) for agentic workflows. The Context API is designed to be integrated into agentic workflows as a core knowledge layer, enabling LLMs to verify source data and prevent hallucinations. Code examples demonstrate using the Python SDK to upload files, query context with metadata filters, and manage conversation history. Custom instructions are now supported for fine-tuning assistant behavior, and new input formats (.json, .md, .docx) and output format (JSON) are available. Region control for EU and US is also introduced.