Curata

Product Introduction

1. Definition: Curata is an AI-native knowledge base platform designed as a read-write repository for both AI agents and human collaborators. It serves as a centralized hub where structured, living documentation is automatically generated from live data sources by AI, then curated and annotated by human teams.
2. Core Value Proposition: Curata exists to eliminate the "knowledge gap" between AI agent outputs and human team understanding. It provides a system where AI agents can continuously write, update, and maintain structured documentation from inputs like CRM data, call transcripts, support tickets, and Slack messages, creating a self-compounding knowledge asset that stays perpetually current without manual overhead.

Main Features

1. MCP (Model Context Protocol) Integration: Curata provides native read/write access via the MCP protocol, acting as a universal connector for any AI agent. By configuring a simple JSON endpoint in clients like Claude Code or Cursor, AI agents gain the tools to search_pages, read_page, write_page, and annotate_page directly within the knowledge base. This enables agents to autonomously fetch context and publish their findings, turning isolated agent runs into collaborative knowledge contributions.
2. Kazam Page Engine & Rich Components: Pages within Curata are built using Kazam, an open-source page engine, in a YAML format. This structure supports over 20 rich component types (and a full component reference API), allowing for the creation of complex, interactive, and well-formatted documentation. Agents and humans can generate content that includes more than plain text, incorporating data visualizations, tables, and interactive elements directly from source data.
3. Human-in-the-Loop Annotation & Versioning: To ensure accuracy and provide governance, Curata features a dedicated annotation layer and full version history. Human team members can review AI-generated pages in a browser, add comments, tag sections, and provide feedback via annotations (e.g., annotate_page). This creates a closed feedback loop where human knowledge refines and corrects agent output, while version history maintains a complete audit trail of all changes, blending automation with curation.

Problems Solved

1. Pain Point: Traditional knowledge bases suffer from documentation rot, becoming outdated as soon as source data changes. Manual updates are time-consuming and often forgotten, leading to team misalignment and agents acting on stale information.
2. Target Audience: Engineering teams building autonomous AI workflows, operations and customer success teams managing client knowledge, product managers documenting features from sprint outputs, and any cross-functional group needing a dynamic, machine-readable "single source of truth" for evolving information.
3. Use Cases: An AI agent processes weekly sales calls and automatically writes a "Customer Insights" page in Curata, highlighting trends and objections. A support agent writes a post-mortem for a critical ticket, and the system automatically generates a linked "Solution Pattern" page for future reference. Product managers have agents that sync feature specs from a live database, keeping the official product documentation perpetually synchronized with the codebase.

Unique Advantages

1. Differentiation: Unlike passive document storage (e.g., Notion, Confluence) or simple vector stores for retrieval-augmented generation (RAG), Curata is a read-write endpoint. It doesn't just retrieve knowledge; it actively structures, writes, and updates it via AI. It also surpasses basic API storage by adding a vital human annotation and review layer, blending AI automation with human oversight in a single workflow.
2. Key Innovation: The core innovation is the synthesis of an MCP-native architecture with a structured, component-rich page format. This combination allows any compliant AI agent to not only query a knowledge base but to contribute to it in a standardized, high-fidelity way. The compounding nature—where each agent run builds upon the existing, versioned knowledge—creates an exponentially more valuable asset over time.

Frequently Asked Questions (FAQ)

1. What is an MCP server and how does Curata use it? An MCP (Model Context Protocol) server is a standardized interface that allows AI agents to interact with external tools and data sources. Curata operates as an MCP server, providing tools like read_page and write_page. This allows any MCP-compatible AI client, such as Claude Desktop or Cursor, to directly read from or write structured content to your Curata knowledge base using a simple JSON configuration and an API key.

2. How does Curata ensure AI-generated content is accurate? Curata implements a human-in-the-loop workflow. While AI agents draft and structure pages from source data, all content resides in a browser-accessible environment where human team members can review, annotate, suggest edits, and approve changes. The version history tracks all modifications, ensuring accountability and allowing teams to correct AI outputs, which then inform future agent runs.

3. Is my data secure if I connect my CRM or Slack to Curata? Curata operates on a principle of secure, controlled access. When you connect external data, the AI agent processes that information to generate structured pages. The knowledge base itself is secured behind API key authentication (for MCP access) and user accounts for browser access. Data flows into Curata to build documentation, but you maintain control over the API keys and who on your team has read/write permissions to the resulting knowledge store.

4. Can I use Curata without any coding knowledge? Yes, the core knowledge base is accessible via a browser for reading and annotating. However, to leverage the automated AI writing features, initial setup requires generating an API key from your Curata settings and configuring an AI tool (like Claude or Cursor) with the provided MCP server URL—a process that involves copying and pasting a configuration snippet. No deep coding is needed, but basic comfort with editing a JSON config file is helpful for setup.