Skip to content

MCP Prompts

AI Summary

MCP Prompts are reusable analysis templates that clients can invoke. Unlike tools (which execute code and return data), prompts return structured instructions for the LLM to follow. Each prompt focuses on a specific domain: draft analysis, hero performance, death patterns, laning, teamfights, itemization, and game turning points.

Overview

MCP Prompts provide pre-built analysis templates for common Dota 2 coaching scenarios. Each prompt:

  • Returns structured instructions for the LLM to follow
  • Focuses on one domain - no scope creep into unrelated analysis
  • Includes benchmarks and frameworks specific to Dota 2

Prompts vs Tools vs Sampling

Feature Purpose When to Use
Tools Execute code, return data Get match data, hero stats, fights
Prompts Guide LLM analysis Structure analysis of provided data
Sampling Server-requested LLM calls Automatic coaching in tool responses

Available Prompts

analyze_draft

Analyzes a draft for lane matchups, synergies, and counter picks.

Parameters:

Parameter Type Required Description
radiant_picks string Yes Radiant's picked heroes
dire_picks string Yes Dire's picked heroes
radiant_bans string No Radiant's banned heroes
dire_bans string No Dire's banned heroes

Analysis Sections:

  1. Lane Matchups - Who wins each lane (safelane, mid, offlane) based on hero matchups
  2. Draft Synergies - How heroes complement each other per team
  3. Counter Picks - Which picks counter enemy heroes and how
  4. Draft Weaknesses - What each draft lacks
  5. Ban Analysis - Why heroes were banned (if bans provided)
  6. Draft Grade - A/B/C/D rating with justification

Focused Analysis

This prompt explicitly excludes teamfight combo analysis and item timing predictions. Those belong in analyze_teamfight and compare_itemization respectively.

Example Usage: 

/dota2-match-analysis:analyze_draft
radiant_picks: "Juggernaut, Shadow Fiend, Earthshaker, Shadow Demon, Pugna"
dire_picks: "Medusa, Pangolier, Magnus, Naga Siren, Disruptor"
radiant_bans: "Chen, Sand King, Enchantress"
dire_bans: "Bane, Centaur, Anti-Mage"

review_hero_performance

Reviews a player's performance with position-specific coaching feedback.

Parameters:

Parameter Type Required Description
hero string Yes Hero played
position string Yes Position (1-5)
kda string Yes K/D/A stats
cs_at_10 string No CS at 10 minutes
gpm string No Gold per minute
item_timings string No Key item timings
fight_participation string No Fight participation %

Analysis Sections:

  1. Overall Rating - Poor / Acceptable / Good / Excellent
  2. Benchmark Comparison - Stats vs position expectations
  3. Key Issues - What went wrong (2-3 points)
  4. Actionable Improvements - What to do differently (2-3 points)
  5. Positives - What was done well (1-2 points)

Position Benchmarks:

Position CS@10 Target Primary Metrics
Pos 1 65-80 CS, GPM, item timings
Pos 2 60-75 CS, kill participation
Pos 3 N/A Space creation, not farm
Pos 4/5 N/A Impact, not net worth

analyze_deaths

Analyzes death patterns to identify preventable mistakes.

Parameters:

Parameter Type Required Description
deaths_list string Yes List of deaths with times/details
focus_hero string No Hero to focus analysis on

5-Question Framework:

Each death is evaluated against:

  1. Was Vision Available? - Ward coverage, minimap awareness, smoke gank?
  2. Power Spike Timing? - Before/after level 6? Key item? Cooldowns?
  3. Objective Trade? - Did death enable Roshan/towers/space?
  4. Buyback Usage? - Available? Used correctly? Wasted?
  5. Item Timing Impact? - How much did death delay key items?

Analysis Sections:

  1. Most Impactful Deaths - Which deaths mattered most (2-3)
  2. Death Patterns - Repeated issues, same area, same killer
  3. Death Categories - Unavoidable / Preventable / Acceptable / Throw
  4. Actionable Advice - How to reduce preventable deaths

review_laning_phase

Reviews the laning phase (0-10 minutes) with coaching feedback.

Parameters:

Parameter Type Required Description
lane_results string Yes Lane winner summary
hero_cs_stats string Yes CS stats at 10 min for all heroes

CS Benchmarks:

Position Poor Acceptable Good Excellent
Pos 1 <50 50-65 65-80 80+
Pos 2 <45 45-60 60-75 75+
Pos 3 Judged by space creation, not CS

Analysis Sections:

  1. Lane Outcomes - Who won each lane and why
  2. Critical Lane - Which lane mattered most
  3. Benchmark Performance - Which cores hit targets
  4. Key Mistakes - What swung lanes
  5. Improvement Advice - What losing lanes could do differently

analyze_teamfight

Analyzes a specific teamfight for coaching insights.

Parameters:

Parameter Type Required Description
fight_time string Yes Game time of fight
participants string Yes Heroes involved
death_sequence string Yes Order of deaths
fight_winner string No Which team won

Analysis Sections:

  1. Initiation - Who started, planned vs reactive, timing
  2. Target Priority - Were right heroes focused?
  3. Ability Usage - Key ults, CC chaining, saves
  4. Positioning - Carries safe? Frontliners absorbing?
  5. Trade Value - What each team gained/lost
  6. Coaching Points - What each side could improve

compare_itemization

Analyzes item choices and suggests alternatives.

Parameters:

Parameter Type Required Description
hero string Yes Hero being analyzed
items_built string Yes Items the player built
enemy_heroes string Yes Enemy team composition
game_context string No Additional context (ahead/behind, etc.)

Analysis Sections:

  1. Build Assessment - Is build appropriate for this game?
  2. Timing Analysis - Were items completed at good times?
  3. Alternative Items - What could have been better?
  4. Item Ordering - Was build order optimal?
  5. Situational Purchases - BKB timing, detection, defensive items

game_turning_points

Identifies and analyzes match turning points.

Parameters:

Parameter Type Required Description
gold_lead_timeline string Yes Gold lead changes over time
key_events string Yes Major events (fights, objectives)

Analysis Sections:

  1. Major Turning Points - 2-4 moments that swung the game
  2. Momentum Shifts - How each turning point changed trajectory
  3. Preventable Swings - Which could have been avoided
  4. Decisive Moment - Single most important moment
  5. Lessons - What each team should learn

Usage with Claude Code

Prompts appear in Claude Code's autocomplete when you type /:

/dota2-match-analysis:analyze_draft
/dota2-match-analysis:review_hero_performance
/dota2-match-analysis:analyze_deaths
...

The prompt will guide Claude to focus on the specific analysis domain without scope creep.

Usage with Other Clients

MCP prompts are exposed via the MCP protocol's prompt listing. Clients can:

  1. List available prompts: list_prompts()
  2. Get prompt template: get_prompt(name, arguments)
  3. Fill in arguments and send to LLM
# Example with MCP SDK
prompts = await session.list_prompts()
draft_prompt = await session.get_prompt(
    "analyze_draft",
    arguments={
        "radiant_picks": "Juggernaut, Shadow Fiend",
        "dire_picks": "Medusa, Pangolier"
    }
)
# draft_prompt contains the filled template to send to your LLM