poe_mcp_suite

mcp
Guvenlik Denetimi
Uyari
Health Uyari
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Gecti
  • Code scan — Scanned 5 files during light audit, no dangerous patterns found
Permissions Gecti
  • Permissions — No dangerous permissions requested

Bu listing icin henuz AI raporu yok.

SUMMARY

Path of Exile MCP Suite

README.md

poe_mcp_suite

This product is not affiliated with or endorsed by Grinding Gear Games in any way.

A suite of Model Context Protocol (MCP) servers that give an AI agent deep, live integration with Path of Exile — from build theory-crafting and passive tree simulation to trade, loot filters, and wiki lookups. It works with any MCP-compatible agent, and ships a workflow framework additionally optimized for Claude Code.

Each server runs independently and exposes a set of tools the agent can call during conversation. Together they let the agent act as an informed PoE assistant: loading your actual build in Path of Building, simulating node choices, checking prices on poe.ninja, searching trade, scoring individual items, and more — all without leaving the chat.

poe_mcp_suite is a master git repo that acts a wrapper for four independently developed MCP tools and some other useful utilities.

Point your AI agent at the AGENTS.md file (or CLAUDE.md if you use Claude Code), and it should find everything it needs to install, connect to Path of Exile and Path of Building, and use all the various API calls for the tools.

If you have any issues or suggestions feel free to e-mail at [email protected], or make changes in your own repos and create pull requests back to this one.


Tool Reference

TOOLS.md — top-level index of all ~172 tools organized by category, with cross-server views (build management, passive tree, jewel awareness, trade, economy, etc.).

For per-server detail (descriptions, arguments), see each submodule's own doc:


Legal & Attribution

This suite uses, processes, and in some cases redistributes data that originated in Path of Exile (© Grinding Gear Games). Before contributing or publishing forks, please read legal_considerations.md — it documents the architectural choices made to minimize redistribution, the community-tool precedent we rely on, and the compliance posture (we will take things down promptly if GGG objects).

Short version: we follow GGG's lead by publishing only the kind of structural tree data they already publish themselves; anything richer (stat description templates, jewel transformation tables, mod pools) is extracted from the user's local install at runtime and never persisted to a public repo. The suite is unaffiliated with and not endorsed by Grinding Gear Games.


Credits

Grinding Gear Games
This project would not exist without Path of Exile — the game itself, the ongoing development, and GGG's decision to publish official passive tree and atlas tree data as open repositories that the community can fork and build on.

Path of Building Community
The PathOfBuilding submodule is a fork of Path of Building Community, originally created by David Gowor and maintained by the PoB Community team.

ianderse/pob-mcp
The pob-mcp server started as a fork of ianderse's pob-mcp project.

boschzilla / poe-trade-mcp
The poe-trade-mcp started as a fork of boschzilla's poe-trade-mcp project.

shalayiding / poe-data-mcp
The poe-data-mcp server started as a fork of shalayiding's poe-data-mcp project.

Craft of Exile
The crafting mod lookup tools use data from craftofexile.com, a community crafting calculator maintained independently of GGG. Craft of Exile is the most comprehensive source of mod pool, fossil affinity, and crafting odds data available to the community. Each user fetches their own copy of the data directly from the site; nothing is redistributed by this project.


A note on your AI model's Path of Exile knowledge

An AI model's built-in PoE knowledge is frozen at its training cutoff — for current models, somewhere between about August 2024 and January 2026 — so content, balance changes, and mechanics introduced after that point may be missing or wrong. Before relying on the model's game-mechanics advice, confirm it against current sources.

The MCP servers exist partly to bridge this gap: use fetch_wiki_page (poe-data-mcp) for up-to-date item and passive descriptions, ninja_lookup / currency_overview for current prices, and parse_pob or the live PoB TCP connection for accurate calc results. When the model's training intuition conflicts with a live tool result, trust the tool.

Training cutoffs for the Claude models the suite is tuned for:

Haiku 4.5 - January 2025
Sonnet 4.6 - Settlers of Kalguur (3.25) - August 2025
Opus 4.8 - Secrets of the Atlas (3.26) - January 2026
Fable 5 - Secrets of the Atlas (3.26) - January 2026


How the agent organizes the work (and how you help)

Path of Exile is a huge game — thousands of items, modifiers, passive nodes, atlas mechanics, league-specific systems, all interacting in ways that change every patch. No single source — not the model's training, not the wiki, not poe.ninja, not Path of Building — is correct about all of it all the time. To make this manageable, the suite uses a small workflow framework that you'll see the agent reference during sessions.

Playbooks

When you start a recognizable task — "help me improve my DPS," "what map mods should I run," "is this rare worth keeping?" — the agent loads a matching playbook from playbooks/. Each playbook contains:

  • A short triage questionnaire to scope the work (goal, budget, locked items, etc.) — answering 3-4 questions up front saves a lot of back-and-forth later
  • A data-load matrix that determines which sources the agent needs (skip the atlas tree if we're doing a gear-only analysis, etc.)
  • The analysis pattern to follow, in order
  • A pitfalls section with concrete lessons from previous sessions ("Diamond Shrine does NOT grant ailment immunity" / "Body armour Eldritch has no mana cost mod") — so prior mistakes don't get repeated

Playbooks are committed to this repo; if a task shape recurs and a playbook doesn't exist yet, the agent can draft one mid-session.

Sharing playbooks back: if you or the agent draft a new playbook that works well for you, or you have improved an existing one with new data sources, or you've added MCPs or tools, please consider opening a pull request to this repo so others can gain from your insights. See playbooks/README.md for format conventions and submission guidelines. Pitfalls discovered in one session can save hours for the next person who hits the same wall.

Reference data cache

Slowly-changing game data — the official GGG passive tree exports, atlas tree, Eldritch implicit pools, shrine mechanics — gets cached locally in reference_data/ so the agent doesn't re-fetch it every session. The directory itself is gitignored (the data is regenerable and large), but reference_data/README.md is committed and tells a fresh clone how to populate it. Each cached file has a fetched: date and league: in its frontmatter so staleness is easy to spot.

Per-character analyses

Detailed build analyses for each character live in character_data/<Account>/<League>/<Character>/ — also gitignored (junction to %APPDATA%/poe_claude_data/ on Windows), since they include playstyle notes and personal info. The agent reads the doc at the start of a session and appends new findings, decisions, and crafting outcomes as you work.

Your role as co-pilot on data hygiene

The agent will narrate what it's doing so you can follow along — "Pulling current Eldritch pool from poewiki — cached version is 14 days old," "Writing this to reference_data/X.md so we don't refetch next time." When data is stale, missing, or possibly out of date for the current league, the agent will ask you for help:

  • "Can you paste an Eldritch boots mod list from your stash so I have current data?"
  • "Has GGG changed how rage decay works recently? My training says X but I want to verify."
  • "What does your current Atlas tree look like? — I don't have an authoritative source for the current league."

Your in-game observations are the single most authoritative source for your specific situation, and your help refreshing the local cache compounds — better data this session means faster, more accurate analyses in every future session. The game is constantly changing; you're the freshest data source we have.

Also - the agent (or any AI) is forgetful after a context compression. I have tried to create a playbook and context-management framework to encourage the agent to always cache the most up-to-date data before doing analyses, and have given it awareness of it's own context so it can intelligently load data and write out any analysis results before compression happens. But the agent still will forget and fall-back to using it's trained-in game data, which is a couple years out of date. Diligence on asking it if it has read the playbooks, loaded recent data sources, and written it's results to the character cache will help keep it on track and provide much better analyses.


Servers

pob-mcp

Repo: charleslucas/pob-mcp · ~123 tools

The core build-analysis server. Connects the agent to Path of Building — either headless (for batch work) or live via a TCP socket to the running PoB GUI. When connected live, every change the agent makes (adding a gem, allocating a passive, swapping an item) appears in the PoB window in real time.

Key capabilities:

  • Load, save, and compare build files
  • Read and write the passive tree, gems, items, config, and notes
  • Simulate passive node upgrades and mastery effect choices by actually running PoB's calc engine — no guessing
  • Validate builds against endgame boss thresholds, check resistances, identify flask immunity gaps
  • Find best anointments, suggest Watcher's Eye mods, rank cluster jewel notables
  • Import characters directly from the PoE API
  • Search trade for upgrades — returns a clickable trade URL (ExileExchange pattern, no automated listing fetches), generate weighted trade queries, check poe.ninja prices

poe-trade-mcp

Repo: charleslucas/poe-trade-mcp · ~33 tools

A multi-server bundle covering the player-facing side of the economy and account data.

Key capabilities:

  • Market: local price-history database with risers/fallers/movers tracking
  • Stash: ⚠️ bulk stash scanning currently blocked — GGG disabled the legacy stash endpoint; OAuth developer registration required for full access. score_rare (individual item scoring) works without auth. See playbooks/stash-scanning.md
  • Trade: search the official PoE trade site with full filter support, look up stat IDs, fetch listing details
  • Character: fetch live gear including all flask slots, ring slots (Ring1–3), and weapon swap slots from the PoE API; export as PoB XML. All required modules (poe_lib, stash_cache, rare_scorer, poe_oauth) are bundled — no external sibling repositories needed.
  • Pricer: price individual or batch items using poe.ninja and a rare-item scorer
  • Filter: read and edit loot filter files in place — find blocks, add/remove/replace rules, set BaseType priorities

poe-data-mcp

Repo: charleslucas/poe-data-mcp · 17 tools

A knowledge and economy lookup server backed by the PoE wiki, poe.ninja, and Craft of Exile.

Key capabilities:

  • Search and retrieve detailed data for gems, unique items, and passive nodes (keystones, notables, masteries, ascendancy)
  • Look up item modifiers (prefix/suffix) by item type
  • Search maps and scarabs with category filtering
  • Current poe.ninja prices and currency exchange rates
  • Fetch cleaned content from any poewiki.net page
  • Parse a Path of Building export code or share URL into a readable build summary
  • Crafting mod lookup via Craft of Exile — search what mods can roll on a base, check fossil/essence affinities, look up base item drop levels. Used as a cross-reference when evaluating whether to craft or buy an upgrade.

PathOfBuilding (api-stdio branch)

Repo: charleslucas/PathOfBuilding

A fork of Path of Building Community with a JSON-RPC API layer added (src/API/). This is the calculation backend that pob-mcp connects to. It exposes PoB's full calc engine — passive tree simulation, item stat computation, full DPS/EHP calculations — over a TCP socket (live GUI mode) or stdio (headless mode).

The API additions live on the api-stdio branch and are designed to be non-destructive: the PoB GUI remains fully usable while the agent works with the same build through the API.


Data Submodules

In addition to the four MCP server submodules above, two data repositories live under reference_data/ as submodules. These are community forks of GGG's official tree exports, augmented with patches overlay files for corrections that GGG hasn't yet published.

poe-skilltree-export (community fork)

Repo: charleslucas/poe-skilltree-export · Path: reference_data/skilltree/

Fork of GGG's official passive skill tree export. The upstream data.json is preserved verbatim; a sibling data_patches.json carries verified corrections with provenance metadata (who verified, when, against what source). Tools merge the two files at load time; on-disk, the GGG export stays pristine so git merge upstream/master works cleanly. See the fork's PATCHES.md for the protocol.

poe-atlastree-export (community fork)

Repo: charleslucas/poe-atlastree-export · Path: reference_data/atlastree/

Same convention applied to GGG's atlas tree export. Identical patch format and merge policy so a single tool can serve both.

Both forks are real submodules and travel with the suite — git clone --recurse-submodules gets you everything. Each fork has origin pointing at the community fork and upstream pointing at GGG's repo, so pulling new GGG releases is a standard git fetch upstream && git merge upstream/master inside the submodule.


Prerequisites

Before cloning, make sure the following are installed and on your system PATH:

Requirement Version Used by Download
Node.js 18 LTS+ pob-mcp nodejs.org
Python 3.10+ poe-trade-mcp, poe-data-mcp python.org
Git 2.x+ submodule checkout git-scm.com
Path of Building Community latest TCP live mode GitHub releases

Windows users: after installing Node.js, restart any open terminals (and VS Code / Claude Code) so the updated PATH takes effect. If Claude Code can't start the pob MCP server, this is almost always the cause.


Quick Start

git clone --recurse-submodules https://github.com/charleslucas/poe_mcp_suite.git
cd poe_mcp_suite

Or if you already cloned without --recurse-submodules:

git submodule update --init --recursive

Install Python dependencies (covers poe-trade-mcp, poe-data-mcp, and RePoE):

pip install -r requirements.txt
pip install -e poe-data-mcp/

Install pob-mcp (Node.js):

cd pob-mcp && npm install && npm run build && cd ..

To use pob-mcp in live mode, launch Path of Building via pob-mcp/LaunchPoBWithAPI.bat rather than the normal shortcut. This sets the environment variables that activate the TCP API server inside PoB.


MCP Client Configuration

1. Copy the template

A .mcp.json.example is included at the repo root. Copy it to .mcp.json and fill in the four CHANGE_ME values:

cp .mcp.json.example .mcp.json
# then open .mcp.json and edit it

The four things to change:

Placeholder What to put
/CHANGE_ME/poe_mcp_suite (×3) Absolute path to your clone, e.g. C:/Users/you/tools/poe_mcp_suite
C:/Users/CHANGE_ME/AppData/... Your PoB builds folder — the Windows default is shown; adjust username or path if needed
CHANGE_ME_your_POESESSID_cookie Your POESESSID session cookie (see below)
CHANGE_ME_YourAccount#1234 Your PoE account name including the # discriminator

Security: .mcp.json contains your session cookie — treat it like a password. It is listed in .gitignore and must never be committed.

2. Get your POESESSID cookie

POE_SESSION_ID is required for importing private characters and running weighted trade queries. To find it:

  1. Log in to pathofexile.com
  2. Open browser DevTools → Application (Chrome) or Storage (Firefox) → Cookies → pathofexile.com
  3. Copy the value of the POESESSID cookie

3. Where the config goes (by client)

The same mcpServers block works across MCP clients — only the file location differs. Two common examples:

  • Claude Code.mcp.json goes in the workspace root (already where you put it above)
  • Claude Desktop — paste the same mcpServers block into:
    • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
    • Windows: %APPDATA%\Claude\claude_desktop_config.json

Other clients (Cursor, Windsurf, Cline, VS Code, Zed, …) use the same block in their own MCP config location.

Yorumlar (0)

Sonuc bulunamadi