deepwa7er

poe2-mcp

Community deepwa7er
Updated

Path of Exile 2 MCP server for build analysis

poe2-mcp

An MCP server for Path of Exile 2 build analysis. Load a build from a Path of Building export code and interrogate it with natural language via any MCP-compatible client (Claude Desktop, Claude Code, etc.).

What it does

  • Decodes PoB export codes (the long base64 string from File → Share → Copy code) or pobb.in share links
  • Exposes stats, passives, items, and skills as MCP tools
  • Searches the full PoE2 passive skill tree, including nodes not yet allocated
  • Filters tree searches to a specific class region using a Voronoi partition — so search_tree("life", classes=["Monk"]) returns only nodes in the Monk area rather than all 4,700 nodes in the tree

Installation

Requires Python 3.11+ and uv.

git clone https://github.com/deepwa7er/poe2-mcp.git
cd poe2-mcp
uv sync

Claude Desktop setup

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "poe2": {
      "command": "uv",
      "args": ["--directory", "/path/to/poe2-mcp", "run", "poe2-mcp"]
    }
  }
}

Claude Code setup

claude mcp add poe2 -- uv --directory /path/to/poe2-mcp run poe2-mcp

Tools

Tool Description
load_build Load a PoB export code or pobb.in URL
get_stats Character stats — life, mana, DPS, resistances, etc.
get_passives Allocated passive nodes with names and stat descriptions
get_items Equipped items and their mods
get_skills Skill socket groups and gem links
search_passives Keyword search within allocated passives
search_tree Keyword search across the full passive tree
get_reachable_nodes Unallocated nodes reachable within N steps from the current build

Class region filtering

search_tree and get_reachable_nodes accept a classes parameter that restricts results to nodes in the specified class territories. This dramatically reduces noise when exploring the tree.

Valid class names: Druid, Huntress, Mercenary, Monk, Sorceress, Warrior

search_tree("life", classes=["Monk"])
search_tree("lightning", classes=["Monk", "Sorceress"])
get_reachable_nodes(max_distance=5, classes=["Monk"])

The partition is computed at startup via a simultaneous multi-source BFS from all six class starting nodes. Each node is assigned to whichever class start is fewest steps away in the tree graph.

Tree data

The passive tree data is sourced from the community-maintained passive-skill-tree-json repository, which mirrors the format published by Grinding Gear Games. The bundled data/poe2_tree.json can be replaced with a newer version by dropping a fresh file in the same location, or by pointing the TREE_DATA_PATH environment variable at an alternative path.

MCP Server · Populars

MCP Server · New

    heymrun

    Heym

    Self-hosted AI workflow automation platform with visual canvas, agents, RAG, HITL, MCP, and observability in one runtime.

    Community heymrun
    Wide-Moat

    Open Computer Use

    MCP server that gives any LLM its own computer — managed Docker workspaces with live browser, terminal, code execution, document skills, and autonomous sub-agents. Self-hosted, open-source, pluggable into any model.

    Community Wide-Moat
    uarlouski

    🚀 TestRail MCP Server

    AI-native MCP server connecting Claude, Cursor, Windsurf, and other AI assistants to TestRail — manage test cases, runs, and results through natural-language conversation, with typed schemas built for LLMs.

    Community uarlouski
    metabase

    Metabase MCP Server

    The easy-to-use open source Business Intelligence and Embedded Analytics tool that lets everyone work with data :bar_chart:

    Community metabase
    kao273183

    MK QA Master

    AI 測試大師 — MCP server driving pytest / Jest / Cypress / Go / Maestro. Analyze, generate, run, advise. Web + Mobile (iOS/Android/BlueStacks).

    Community kao273183