MCP Dota 2 Match Analysis Server¶

AI Summary

MCP server for Dota 2 match analysis. Run with uvx (uvx dota2-match-analysis@latest) or Docker (docker run --pull=always dbcjuanma/mcp_replay_dota2). Tools (LLM calls these): get_match_heroes, get_match_players, get_hero_deaths, get_combat_log, get_fight_combat_log, get_item_purchases, get_objective_kills, get_match_timeline, get_stats_at_minute, get_courier_kills, plus pro scene tools. Game State Tools: list_fights, get_teamfights, get_fight, get_lane_summary, get_cs_at_minute, get_snapshot_at_time, get_farming_pattern. Resources (static context): dota2://heroes/all, dota2://map, dota2://pro/players, dota2://pro/teams. Many tools are parallel-safe for efficient multi-point analysis. Connects to Claude Desktop, Claude Code, LangChain, LangGraph, CrewAI, or direct API integration.

A Model Context Protocol (MCP) server that gives LLMs the ability to analyze Dota 2 matches by parsing replay files and querying the OpenDota API.

Coaching Philosophy¶

This server includes built-in coaching instructions that guide LLM responses to provide meaningful analysis, not just raw data dumps:

Pattern Recognition: Identify repetitive death patterns (support positioning, smoke ganks, vision-related deaths)
Objective Analysis: Link tower kills to hero rotations (mid with rune, supports via portal, defender TPs)
Economy Tracking: Explain networth swings in context of teamfights and objectives taken
Actionable Advice: Every analysis ends with specific, actionable coaching points

The LLM is instructed to analyze the "why" behind the data, connecting statistics to game-changing decisions.

What This Does¶

This server exposes tools and resources that an LLM can call to answer questions about Dota 2 matches:

"Why did we lose the teamfight at 25 minutes?"
"How did the enemy Anti-Mage get such a fast Battle Fury?"
"When did Roshan die and who took the Aegis?"
"Show me what happened when I died at minute 12"

The LLM reads the replay data and provides analysis based on actual game events, not guesswork.

How MCP Works¶

┌─────────────────┐         ┌─────────────────┐         ┌─────────────────┐
│   LLM Client    │ ──MCP── │   This Server   │ ──────▸ │  Replay Parser  │
│ (Claude, GPT)   │         │                 │         │  OpenDota API   │
└─────────────────┘         └─────────────────┘         └─────────────────┘
         │                           │
         │   "Analyze match 123"     │
         │ ─────────────────────────▸│
         │                           │
         │                           │── calls get_hero_deaths(123)
         │                           │── calls get_combat_log(123, ...)
         │                           │── calls get_objective_kills(123)
         │                           │
         │   structured JSON data    │
         │ ◂─────────────────────────│
         │                           │
         ▼                           ▼
   LLM synthesizes response: "The fight was lost because..."

Resources = Static data the LLM can reference (hero list, map positions) Tools = Functions the LLM can call with parameters (get deaths, get combat log)

Quick Start¶

Option A: uvx (Recommended)¶

uvx dota2-match-analysis@latest

The @latest suffix ensures you always get the most recent version.

Option B: Docker¶

docker run --pull=always -p 8081:8081 dbcjuanma/mcp_replay_dota2 --transport sse

The --pull=always flag ensures you always get the most recent image. Connect to http://localhost:8081/sse. See Docker Guide for details.

Option C: From Source (Contributors)¶

git clone https://github.com/DeepBlueCoding/mcp-replay-dota2.git
cd mcp-replay-dota2
uv sync
uv run python dota_match_mcp_server.py

2. Connect to Your LLM¶

See Integrations for setup with:

Claude Desktop
Claude Code CLI
OpenAI + LangChain
Custom Python clients

3. Ask Questions¶

Once connected, just ask naturally:

"Analyze match 8461956309. Why did Radiant lose?"

The LLM will automatically call the appropriate tools and synthesize an analysis.

Available Tools¶

Match Analysis¶

Tool	What It Does
`download_replay`	Pre-cache replay file (call first for new matches)
`get_match_info`	Match metadata (teams, players, winner)
`get_match_heroes`	10 heroes with KDA, items, stats, counter picks
`get_match_players`	10 players with names and hero assignments
`get_match_draft`	Complete draft order (bans/picks)
`get_match_timeline`	Net worth, XP, KDA over time
`get_stats_at_minute`	Snapshot of all players at a specific minute ⚡
`get_hero_deaths`	All deaths with killer, victim, ability used
`get_combat_log`	Damage events, abilities, modifiers in a time range
`get_fight_combat_log`	Auto-detects fight boundaries around a death
`get_item_purchases`	When each item was bought
`get_objective_kills`	Roshan, towers, barracks timings
`get_courier_kills`	Courier snipes with position
`get_rune_pickups`	Rune pickups by hero

Game State Analysis¶

Tool	What It Does
`list_fights`	All fights with teamfight/skirmish classification
`get_teamfights`	Major teamfights (3+ deaths)
`get_fight`	Detailed fight info with positions ⚡
`get_camp_stacks`	Neutral camp stacking events
`get_jungle_summary`	Stacking efficiency by hero
`get_lane_summary`	Laning phase winners and hero stats
`get_cs_at_minute`	CS/gold/level at specific minute ⚡
`get_hero_positions`	Hero positions at specific minute ⚡
`get_snapshot_at_time`	High-resolution game state at specific time ⚡
`get_position_timeline`	Hero movement over time range ⚡
`get_fight_replay`	High-resolution replay data for fights ⚡
`get_farming_pattern`	Analyze farming pattern (creeps, camps, movement)

⚡ = Parallel-safe: Call multiple times with different parameters in parallel for faster analysis.

Pro Scene¶

Tool	What It Does
`search_pro_player`	Fuzzy search for pro players
`search_team`	Fuzzy search for teams
`get_pro_player`	Get player details by account ID
`get_team`	Get team details + roster
`get_team_matches`	Team match history with series grouping
`get_leagues`	All leagues/tournaments
`get_pro_matches`	Recent pro matches with series grouping (supports head-to-head filtering)
`get_league_matches`	Matches from a specific league

Available Resources¶

Static reference data (user attaches to context):

URI	Data
`dota2://heroes/all`	All 124 heroes with attributes
`dota2://map`	Tower, camp, rune, landmark positions
`dota2://pro/players`	All pro players
`dota2://pro/teams`	All pro teams

Match-specific data uses tools

For match heroes/players and detailed pro player/team info, use the corresponding tools (get_match_heroes, get_match_players, get_pro_player, get_team).

Example Conversations¶

See Use Cases for detailed examples:

Analyzing why a teamfight was lost
Tracking a carry's item timings
Understanding a gank that went wrong
Comparing laning phase performance