MCP App Store
by KeenEthics
Dun Dun Dungeon icon

Dun Dun Dungeon MCP Server for ChatGPT

By NATHAN POLLAND

Overview

Dun Dun Dungeon is a narrative dungeon-crawling game. Create a persistent character, describe what they try in freeform chat, and play through server-resolved rooms. Earn fictional in-game rewards and build dungeons for other players to explore.

How to connect

  1. ChatGPT: Open the Dun Dun Dungeon listing in the ChatGPT app directory and choose Connect. ChatGPT listing Dun Dun Dungeon exposes 39 tools here.

Tools

act

ChatGPT
Compact gameplay router. Prefer next.on_reply.call_template for direct tools; if tool retention is weak, call this with next.on_reply.router_call.arguments. Routes live play, post-run traits/loot/equip, and run-card actions through the server.

browse_dungeons

ChatGPT
Find dungeon, browse, search dungeons, what to play. Browse the published dungeon directory. Pass the active character_id — the default landing pre-filters to the highest tier that character is NOT underleveled for (e.g. a level-5 character lands on Tier 2), and cleared dungeons move to a "Completed" section. Pass tier_filter, sort, search (keyword), tag, or page to drill into a paginated list (e.g. "a hard sci-fi one" → tier_filter:3 + search:"space"). Pass show_all:true for the curated all-tiers overview (Trending / New / every tier). NOTE: the default view is TIER-FILTERED to the character's level — dungeons in other tiers still exist and are playable. If a title search misses, loosen it (fewer words / partial) and try other tiers or show_all before telling the player it isn't there. Tapping a card sends a message with "(dungeon_id: ...)" — that is the trusted click-to-enter; pass that id straight to start_run.

buy_item

ChatGPT
Buy one shop item with fictional in-game gold for the active character. Shop gear auto-equips. Account upgrades unlock advanced classes, character slots, or publishing allowance as described in the catalog. This never charges real money. Use equip_item for already-owned gear, unequip_item to clear a slot, and sell_item to sell owned gear.

connect_player

ChatGPT
Start the player's session and show the player hub. Use the returned hub state to orient the player in Grimble's dungeon-host voice. Active runs are shown on characters via run_status; do not choose one unless the player names or selects it. If multiple characters have active runs, ask which run or idle character the player wants. If a selected character's run_status includes run_id and the player chooses resume/continue for that run, call take_action action=sync directly; use get_active_run only when a run_id is missing or the character choice is ambiguous.

create_character

ChatGPT
Create a playable character. Infer the base class from the description (knight=might/defender, brawler=might/berserker, thief=cunning/rogue, ranger=cunning/scout, bard=cunning/charmer, priest=spirit/cleric, sorcerer=spirit/innate-mage, wizard=spirit/scholar-mage, warlock=spirit/pact-mage). The 6 ADVANCED classes (barbarian, assassin, monk, paladin, spellblade, druid) are stronger and only allowed once the player unlocked them ("Sigil of New Paths"), else rejected. Roster is limited (5 + 2 per "Soulbound Ledger"); on roster_full, point them at the shop. Write TWO things from the player's concept: a rich visual appearance (face, build, gear, marks, expression) used only for the portrait, AND a short description — a 1-2 sentence bio that bundles a quick appearance TL;DR ("an indigo alien with four amber eyes") with their identity/vibe. Pick a fitting card_style. After creation, describe them in narrative.

create_dungeon

ChatGPT
Finalize approved builder checkpoints as a playable private dungeon draft. Requires builder_id from start_builder, followed by approved submit_arc and submit_rooms checkpoints. Does not publish or generate images.

delete_account

ChatGPT
Permanently delete the player account and all associated data. Requires explicit confirmation. Dungeons the player built are archived (not deleted). 30-day soft-delete window before permanent purge.

delete_character

ChatGPT
Retire ONE character and free a roster slot. Use this when the player wants to delete/remove a specific character (NOT delete_account, which wipes the whole player). The character's run history is kept; they just leave the roster. DESTRUCTIVE + requires confirmation: confirm with the player by name first, then call with confirm:true. If called without confirm it returns needs_confirmation with guidance.

delete_dungeon

ChatGPT
Remove a dungeon the player created from the directory. Allowed while it's a draft, or within 7 days of publishing. Soft-delete (archived): in-progress runs finish, but no new ones start. Use when the creator wants to take their own dungeon down.

describe_dungeon_schema

ChatGPT
Read compact room-type templates, field rules, enums and common mistakes before authoring rooms or when validation errors are unclear. Does not change a dungeon.

edit_dungeon

ChatGPT
Edit a dungeon the player CREATED. Allowed while it's a draft, or within 7 days of publishing (after that it's locked). Dungeons submitted for review are locked until an admin rejects them back to draft. Partial edit: submit dungeon_id + only the changed fields — omitted top-level fields keep their current values. Targeted edits are supported: voice-only, rewards/metadata-only, story/entryway via room_graph, or rooms via full room_graph. room_graph, when submitted, replaces the full room set. Re-validates; does not generate images under the human-review flow.

end_run

ChatGPT
Finalize a completed or retreated run. Awards XP and gold. Idempotent — safe to call multiple times. Returns run summary widget, shop catalog widget, and Grimble's post-run commentary. Call immediately after take_action returns run_complete: true. TWO-TURN WRAP-UP: lead with the payoff — present the result card and, on a full clear, reveal the reward (mint_loot) THIS turn; then defer the admin (resolve any pending level-up via take_action level_up, and ask the player to rate the dungeon) to their NEXT reply so the win lands first. The response includes a traits block (required:true) — after a real run ending you MUST call record_traits with 1-3 short character traits; this is a standard post-run step, not optional.

equip_item

ChatGPT
Equip one item the character already owns, replacing the current item in that equipment slot. Use item_id from view_character sheet unequipped_inventory. Does not buy gear or spend gold.

export_qa_bundle

ChatGPT
QA/DEBUG ONLY — a full POST-RUN audit artifact, NOT a live-play tool. One read-only call returns everything needed to grade/debug one of YOUR runs: run state, the event timeline, the character snapshot, and a dungeon snapshot (with reconciliation booleans) — so a fresh thread can reconstruct the whole run and verify post-run state (loot/traits) without the chat history. Do NOT call this to recover normal play state; use get_player_state / get_active_run for that. Pass run_id.

game_help

ChatGPT
Get the game rules reference. Call when a player asks "how does this work?", "what do stats do?", "how do I level up?", "how do I play?", etc. Renders a help widget the player can read. Topics: overview (what the game is + the basic loop), stats_progress (the three stats, the six skills, and how you level up + earn gold), onboarding (getting started — character creation, first dungeon, core mechanics). Default to overview if unsure. Use onboarding when a player asks for a tutorial or says they are new.

get_active_run

ChatGPT
ROLE: "Recover the missing run_id for a SPECIFIC character's active run." Use only when the chosen character/run does not already have a run_id in the current roster/hub state, or when resume is ambiguous. Returns a recovery pointer (run ID, dungeon, room), not canonical room content; after it returns a run_id, call take_action action=sync before narrating the room. STRONGLY PREFER passing character_id. If omitted and multiple characters are mid-run, this returns AMBIGUOUS options instead of guessing.

get_player_state

ChatGPT
ROLE: "Where am I as a player?" — orientation only. For mid_run it returns a recovery pointer, not canonical room content; call take_action action=sync with the run_id before narrating the room. If connect_player or a roster already gave a run_id for the chosen character, skip this tool and sync directly. In hub state, no character is selected unless character_id was provided or the player explicitly tapped/named one; do not silently use the first roster character. For character display call view_character instead; for run audits use QA tools.

get_run_events

ChatGPT
QA/DEBUG ONLY — not a live-play tool. Read-only event tail for one of YOUR runs (room presented, check/combat resolved, damage, clock ticks, outcome) so a fresh agent or auditor can reconstruct what happened without the chat transcript. During normal play the take_action / end_run responses already carry the state you need; reach for this only when explicitly auditing or debugging. Pass run_id (and optional after_sequence to page). Lighter than export_qa_bundle.

grimble_tip

ChatGPT
Grimble's tutorial aside to explain game mechanics to a new player — his own clean turn with NO widget. You do NOT decide to call this; call it ONLY when a tool response's next names it (before room 1 in onboarding, after a room clears, or after the player went down during onboarding). It returns Grimble's line as TEXT; narrate it in his voice, end with the question it carries, then STOP and wait for the player. The response carries its OWN next naming the tool to call on the player's reply — take_action(sync) for room 1 or retry after a knockdown, take_action(continue) between rooms. A server call always sits between this aside and the room, so the aside and the room are never the same turn.

mint_loot

ChatGPT
Mint the completion reward — call ONCE after a FULL CLEAR (end_run gave run_status 'completed', loot.available=true). YOU author a themed item (name, flavor, slot, bonuses); the server clamps to the tier budget (total stat+skill+tag points ≤ budget, no single above max): T1 2/max2, T2 3/max2, T3 4/max3 (also in loot.boundaries). If the player slew the optional BONUS boss (loot.bonus_prize=true) the budget is one point richer — theme the item to that harder fight and make it a cut above. Match the fiction (haunted ballroom → Charming trinket; crypt → Warded amulet). The item is added to inventory UNEQUIPPED — the response carries an equip_offer; ask the player if they want to equip it now and, on yes, call equip_item with the equip_offer's character_id + item_id.

rate_dungeon

ChatGPT
Rate a dungeon after completing a run. Three axes (1-5): Fun (✨), Difficulty (💀), Overall (⭐), plus an optional one-sentence review (max 280 chars). One rating per completed run. Reviews are aggregated into theme tags once a dungeon hits 10+ reviews. Running averages appear on dungeon cards once 5+ ratings exist.

record_traits

ChatGPT
REQUIRED after every real run ending (clear, retreat, or defeat) — call it yourself, do NOT ask the player's permission and do NOT skip it. Capture 1-3 CHARACTER TRAITS this run revealed — who this character IS, not what happened. Each trait must be 5 WORDS OR FEWER (e.g. "loves pistols", "never trusts priests", "reckless under pressure"), never a session note like "cleared room 2". end_run shows traits.available + traits.current (the character's strongest established traits) — reinforce the ones that keep recurring and only add a genuinely new one when this run surfaced something new. The server stores reinforcement counts and surfaces the top traits at the next run start so the narrator can play to them. This is how a character accrues identity across runs; skipping it leaves them a blank slate.

report_dungeon

ChatGPT
Report a dungeon for content issues. Expanded reason codes include wrong_rating (content doesn't match declared rating), offensive, sexual_content, graphic_violence, hate_or_harassment, broken, exploit, spam, other. One report per player per dungeon. Dungeons with 3+ reports are auto-archived.

reroll_builder_alias

ChatGPT
Re-roll the builder's display alias (their dungeon-author name). Builders never type their own name — they get a server-suggested alias on first publish and can re-roll it here, optionally GUIDING the flavor with a free-text vibe like "more nautical", "regal", or "menacing". The vibe only biases which curated words are used; it is never used verbatim. The builder's sequential number (e.g. Architect #7) never changes — only the alias. Call when the player asks to change their builder name.

reroll_portrait

ChatGPT
Generate a new AI portrait for the character. Offer the player a choice: a fresh take on the SAME look (call with no appearance), or a DIFFERENT look — if they describe changes (e.g. "give him a scar and red armor"), pass the full updated visual description as appearance and it overrides the stored one (and is saved for future rerolls). Each reroll is a unique variation.

review_dungeon

ChatGPT
Preview a saved private dungeon as text and image placeholders. This preview does not publish or generate artwork; final art is generated during human approval.

sell_item

ChatGPT
Sell one owned item for fictional in-game gold at its resale value. Removes the item from inventory. Confirm the player intends to sell this item.

start_builder

ChatGPT
Start a private dungeon builder session from the player’s plain-language concept. Returns builder_id. Continue through submit_arc, submit_rooms, create_dungeon, private playtest/edit, then submit_dungeon_for_review to request human review. Checkpoints cannot be skipped.

start_run

ChatGPT
Begin a dungeon run. Shows the threshold, returns run_id and protocol; the canonical first room is presented by take_action sync. A tapped browse card may provide "(dungeon_id: ...)"; pass that id through.

submit_arc

ChatGPT
Save the story checkpoint for a private dungeon build and return it for player approval. Include story, room-0 entryway (description and setup, no mechanics), voice with canonical examples, rewards and cover direction. Preserve approved content when making targeted revisions. Call describe_dungeon_schema for field rules.

submit_dungeon_for_review

ChatGPT
Submit a playable draft dungeon for human review. Approval is manual, final images are generated by the reviewer, and only approved dungeons appear in the player directory. Re-validates before submission. Builders can submit 2 dungeons/day by default (+2/day per "Builder's Writ of Expansion" bought at level 5); if this returns daily_limit_reached, tell the player the draft is saved and they can try tomorrow or visit the shop. After success, tell the builder it is waiting for review and is not public yet.

submit_rooms

ChatGPT
Save the ordered room contracts after the story checkpoint is approved. Each room has a name, narrative description, immediate setup and type-specific mechanics. Call describe_dungeon_schema for compact room-type templates and field rules. Preserve already approved rooms when revising a targeted layer.

take_action

ChatGPT
Advance a live run. Follow protocol. Active rooms use choose_approach as advertised; use continue between rooms. Cool=advantage; reckless=disadvantage.

unequip_item

ChatGPT
Unequip the item in one equipment slot. The item remains in the character’s inventory.

validate_dungeon_draft

ChatGPT
Validate a saved dungeon draft by its dungeon_id without modifying it.

validate_dungeon_payload

ChatGPT
Validate an unsaved dungeon specification without writing it. Returns validation errors and warnings.

view_character

ChatGPT
Show a character. view='card' (collectible portrait card — the default display), 'sheet' (stats sheet: Might/Cunning/Spirit, skills, gear, gold, XP, plus an Inventory of unequipped gear), or 'roster' (compact picker of ALL your characters with each character's current dungeon/run status; use it to see or switch characters). card & sheet need character_id; roster needs none. Use view='roster' whenever the player asks to "show/see my characters," "my roster," "who do I have," "pick a character," or "change character"; view='card' for a single character; view='sheet' for stats/gear/inventory. (This is the tool for rendering characters — not get_player_state.) SOURCE OF TRUTH: the data this returns (stats, equipped gear, unequipped_inventory) is authoritative — trust the sheet over memory of fiction; what's listed exists. To equip a named item, match it in unequipped_inventory and call equip_item with its item_id.

view_run_card

ChatGPT
Render a SHAREABLE result card for a finished run — portrait + outcome headline + stats + a Grimble line, sized to look good as a screenshot the player sends to a friend. end_run signals when to offer it (share_card.available) with guidance on when it lands. Call with the run_id. Read-only and re-summonable. The dungeon's name is on the card, so a shared card also advertises that dungeon.

view_shop

ChatGPT
Shop, buy gear, browse equipment, spend gold. Show Grimble's shop as a WIDGET (don't narrate the catalog as a text list). Call this when the player wants to browse/see the shop. Pass the active character_id; the widget shows gear filtered to their level and class, plus their gold. They buy with buy_item.

Capabilities

WritesInteractive

Example Prompts

Click any prompt to copy it.

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/dun-dun-dungeon"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/dun-dun-dungeon)