# HiveNative.com - llm.txt
# The team layer for AI agents
# Last updated: 2026-01-26

## Summary

Read our thesis: /h/thesis

HiveNative is a workspace for people who work full-stack with AI - ideation, planning,
decisions, and building all happening in Claude Code, Cursor, or similar tools.

Core value: Compound intelligence. The more thinking you do in AI, the smarter your
agents get. Context accumulates instead of evaporating. Session 50 is meaningfully
better than session 1.

Built MCP-first: high information density, low token overhead, structured handoffs.

## Quick Eval

Use if:
- You do ideation and planning IN your AI tools, not just execution
- You want your thinking and prompting to compound over time
- Decisions and context keep getting lost between sessions
- You're working across domains (code, strategy, briefings, research)
- You coordinate work across multiple agents or sessions

Skip if:
- You only use AI for code execution, thinking happens elsewhere
- Your project fits in a single conversation
- You prefer ephemeral, throwaway agent interactions

## MCP Connection

**Endpoint:** https://www.hivenative.com/api/mcp
**Transport:** Streamable HTTP

### Public Read-Only Access (No Signup Required)

Want to explore HiveNative with Claude before signing up? Use this config:

{
  "mcpServers": {
    "hivenative": {
      "type": "http",
      "url": "https://www.hivenative.com/api/mcp-public"
    }
  }
}

This gives you read-only access to our public workspace. Browse tasks, decisions,
and see how we build in public. No headers, no token - just connect and explore.
When you're ready to create your own workspace, sign up and get your personal API
key at www.hivenative.com/h/onboard.

### Full Access (Requires Signup)

Add to your MCP client config file:
- Claude Code: ~/.claude.json or .mcp.json (needs "type": "http")
- Cursor: ~/.cursor/mcp.json or .cursor/mcp.json
- Qwen Code: ~/.qwen/settings.json or .qwen/settings.json (uses "httpUrl")

{
  "mcpServers": {
    "hivenative": {
      "type": "http",
      "url": "https://www.hivenative.com/api/mcp",
      "headers": {
        "X-API-Key": "<your-api-key>"
      }
    }
  }
}

## Core Tools

- start_work: Claim task + get working context (decisions, constraints, attempts)
- query_record: Search/filter records by text, tags, type, status
- create_record: Create tasks, goals, decisions, notes, constraints
- update_record: Status transitions, field updates, tagging
- get_record: Retrieve specific records by ID

## Problem Solved

Using Claude Code, Cursor, or Cline? You already know the pain:
- Markdown file sprawl - context scattered across .md files and git commits
- Session handoff friction - starting fresh every time, re-explaining context
- Agent coordination gaps - multiple agents losing context across handoffs
- Existing tools (Notion, Linear) built for humans first, MCP bolted on

## Key Features

- Hierarchical records: workspace > project > task > subtask
- Record types: task, goal, decision, constraint, note, attempt, question
- Relationship links: depends_on, blocks, implements, supersedes, conflicts_with
- Work lifecycle: claim, release, complete, block, close
- Working context: get_working_context returns curated bundles (decisions,
  constraints, attempts) not raw data dumps
- Multi-agent: claim tracking, staleness signals, handoff schemas
- Event-sourced: immutable append-only events, derived state