Claude MCPBeginner15 min

Claude MCP + HatiData: Full-Stack Agent Setup

Set up Claude Desktop with all 24 HatiData MCP tools for a complete agent-native data experience including memory, CoT, triggers, branching, and queries.

What You'll Build

A complete Claude Desktop setup with all 24 HatiData MCP tools for memory, CoT, triggers, branching, and queries.

Prerequisites

$hati init

$Claude Desktop installed

Architecture

┌──────────────┐  MCP   ┌──────────────┐    ┌──────────────┐
│   Claude     │──────▶│  HatiData    │───▶│   Engine     │
│  Desktop     │       │  MCP Server  │    │  + Vectors   │
└──────────────┘       │  (24 tools)  │    └──────────────┘
  Memory • CoT • Triggers • Branches • SQL

Key Concepts

●24 MCP tools: memory (5), query (4), CoT (3), triggers (4), branches (5), identity (3) — all available in Claude Desktop
●Zero cloud dependency: everything runs locally with hati init — your data never leaves your machine
●Automatic tool selection: Claude chooses the right HatiData tool based on your natural language request
●Persistent state: all data (memories, tables, branches, CoT traces) persists across Claude Desktop sessions

Step-by-Step Implementation

Install HatiData Locally

Install and initialize HatiData with all agent-native features enabled.

Bash

# Install HatiData CLI
curl -fsSL https://hatidata.com/install.sh | sh

# Initialize local instance with all features
hati init --features memory,cot,triggers,branches

# Verify all services are running
hati status

Expected Output

HatiData CLI installed: v1.0.0

Initializing HatiData...
  Query engine: started
  Vector index: started
  Memory layer: enabled
  CoT ledger: enabled
  Semantic triggers: enabled
  Branch engine: enabled

Services:
  SQL Proxy: localhost:5439 (healthy)
  MCP Server: localhost:5440 (healthy)
  Tools available: 24

Note: hati init starts everything locally — no cloud account required. All data stays on your machine.

Configure Claude Desktop

Add HatiData as an MCP server in Claude Desktop's configuration.

JSON

{
  "mcpServers": {
    "hatidata": {
      "command": "hati",
      "args": ["mcp-server"],
      "env": {
        "HATIDATA_HOST": "localhost",
        "HATIDATA_PORT": "5440"
      }
    }
  }
}

Expected Output

Save to:
  macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  Linux: ~/.config/Claude/claude_desktop_config.json

Note: After saving, restart Claude Desktop. The HatiData tools will appear in the tools panel.

Test Memory Tools

Try the memory tools: store_memory, search_memory, get_agent_state, set_agent_state.

text

# In Claude Desktop, try these prompts:

# 1. Store a memory
"Remember that I prefer Python over JavaScript and work at a fintech startup"

# 2. Search memories
"What programming language do I prefer?"

# 3. Set agent state
"Save my preferences: timezone=PST, editor=VS Code, deploy_target=AWS"

# 4. Get agent state
"What are my saved preferences?"

Expected Output

# store_memory → stored with embedding for semantic search
# search_memory → "You prefer Python over JavaScript"
# set_agent_state → preferences saved as JSON
# get_agent_state → {timezone: "PST", editor: "VS Code", deploy_target: "AWS"}

Note: Memory tools: store_memory (5 tools), search_memory, get_agent_state, set_agent_state, delete_memory.

Test Query and CoT Tools

Try the SQL query tools and chain-of-thought logging.

text

# In Claude Desktop, try these prompts:

# 1. SQL query
"Create a table called projects with name, status, and deadline columns,
 then insert 3 sample projects"

# 2. Chain-of-thought
"Start a reasoning session and walk me through how to design
 a microservices architecture for an e-commerce platform"
# (Claude will automatically log each reasoning step to the CoT ledger)

# 3. Replay CoT
"Replay the reasoning session from earlier about microservices"

Expected Output

# query → CREATE TABLE + INSERT executed successfully
# log_reasoning_step → 5 steps logged to CoT ledger
# replay → Full reasoning trace with hash verification

Note: Query tools: query, list_tables, describe_table. CoT tools: log_reasoning_step, replay_decision, get_session_history.

Explore All 24 Tools

Explore branching, triggers, and advanced tools.

text

# In Claude Desktop, try these prompts:

# Branching
"Create a branch called 'experiment', add a test table,
 then merge it back to main"

# Triggers
"Set up a trigger that alerts me when any conversation
 mentions 'production deployment'"

# Full tool list
"List all available HatiData tools and their descriptions"

Expected Output

Available HatiData MCP Tools (24):
  Memory:    store_memory, search_memory, get_agent_state, set_agent_state, delete_memory
  Query:     query, list_tables, describe_table, query_arrow
  CoT:       log_reasoning_step, replay_decision, get_session_history
  Triggers:  register_trigger, list_triggers, delete_trigger, test_trigger
  Branches:  branch_create, branch_query, branch_merge, branch_discard, branch_list
  Identity:  create_api_key, rotate_api_key, list_keys

Note: All 24 tools are available in every Claude Desktop conversation. The MCP server handles routing to the correct HatiData subsystem.

Related Use Case

AI/ML

Multi-Agent Research Engine

Multi-Agent Research with Shared Memory

Ready to build?

Install HatiData locally and start building with Claude MCP in minutes.

Join Waitlist