Skip to main content
Glama

Agentic MCP tools for Sid Meier's Civilization V: Brave New World

Give your agent a civilization. See what survives.

Civilization V armies gathering across a river and contested border, official game screenshot

Game image from Civilization V on Steam.

civ-v-llm-harness MCP server – quality and maintenance score on Glama

Built for agentic play

A rival settles the river first. Your peaceful science plan now has a border problem. The agent in that seat sees the same fogged map you would, knows only the civilizations it has met, and has to decide anyway.

This harness puts an AI agent in a real seat in Sid Meier's Civilization V. You own the game, you host the table, and the agent takes a human seat: across from you in hotseat or over LAN, in your chair against the game's AI, or beside other agents in one match. It plays by the same rules as everyone else at the table and sees no more than you would.

Ready to play? Give an LLM of your choice the agent install guide and ask it to set up the harness and start a game. The guide has the commands and checks; this page is for you. Give your agent the raw guide.

Where the project lives. The one official, write-enabled repository is on GitLab: https://gitlab.com/Tyler-Meador/civ-v-linux-mcp. Issues, merge requests and releases are there. The GitHub repository is a read-only mirror of it, kept for visibility; issues and pull requests are switched off there. To report a bug or send a change, start with CONTRIBUTING.md.

Requirements

Current release

Ways to play

Native Linux Steam build, Brave New World

2.17.0 · Lua runtime v341

Solo · hotseat · LAN · multi-LLM

On this page

Related MCP server: civ6-mcp

What you need

The game

Sid Meier's Civilization V with Brave New World on your own Steam account, as the native Linux build. Steam downloads it once you turn Proton off for the game.

A Linux desktop

x86_64 with a display. The game runs in a normal window on your screen. A Steam Deck works (docs/DECK_HOWTO.md).

An agent that speaks MCP

Claude Code, Codex CLI or any other MCP client, with whatever agent and budget you give it. Each agent at the table gets its own server.

Python 3.11 or newer

No compiler and no uv needed.

Setup is one sitting, and your agent does it. Hand it the install guide and it works through the checks, telling you exactly when it needs a click from you in Steam (switch the game off Proton, enable the tuner in the game's config). The game itself takes about two minutes to cold start.

Fair play

The agent sits in a human seat and gets a human's information, nothing more. That is the whole design.

  • Fog of war is fog. Fogged tiles carry no live units. Unexplored land is a frontier the agent has to go and look at, not a list of unknown tiles to plan over.

  • Unmet civilizations do not exist until the agent meets them. The AI's private plans stay private.

  • Diplomacy goes through the game's own screens. Deals, demands, peace terms, friendship, denouncement and leader conversations are the same ones the AI offers you, with the AI's actual replies. Between human seats (you and the agent, or two agents) everything goes over the trade table, as it does for you.

  • A seat is bound to its channel. With a seat table on the daemon (--seats), each connection is bound to one seat by a token or by its OS user, the daemon writes that seat into every call itself, and a server started for another seat, a seat change onto it, or any raw Lua is refused. Two agents in one hotseat see only their own seat (docs/AGENT_INSTALL.md, "Two agents, one game").

  • Refusals mirror the UI. If the game would not let you click it, the harness refuses it and says why. Every screen, hover and refusal reason a human reads is what the agent reads.

  • No back door. The raw Lua escape hatch is off unless you turn it on. docs/LIMITATIONS.md lists what is withheld and why, each item checked against the engine.

  • The whole game. Units, cities, research, policies, religion, espionage, World Congress, trade routes, archaeology, ideology, great works, city-states and the spaceship. No shortcut to a victory.

Built for how a model plays

The world reaches the agent as text shaped to its way of thinking, so it spends tokens on decisions rather than on scrolling. That is what keeps a turn affordable and the game moving while you wait for it.

  • Tiles as text, at the coordinates the move tools take. Every revealed tile comes back as terrain, yields, resources, what a worker could build there and who is standing on it.

  • The rules once, not on every row. reference is every unit, building, tech, policy, promotion, belief and terrain with its effect text, read once from the game's database (mods included). Lists carry names and live numbers only, so no hover is paid for twice.

  • One call per turn. turn(action="finish") ends the turn, waits, and comes back with the new turn's status, what happened in between, and the agent's own notes. briefing folds a whole turn into one compact read: about 2.5 KB for a 38-unit empire where the separate reads are 30 KB.

  • It sleeps through quiet turns, by its own choice. With skip_quiet_turns the agent asks to be left alone until something a human would look up for: combat of its own, a leader at the door with a decision, an empty city, happiness falling out of the comfortable band, a deal in its last turns (once), a threat beside a unit on the move. Other players' wars, alliances, denouncements and quests, a splash the harness already swept and a leader's plain remark are quiet. The agent picks how many turns to let pass; the harness only ever wakes it early.

  • Standing orders. notebook(action="give_order") hands a unit a short plan (walk there, build a farm; heal to 80%, go back, fortify) that runs itself at the start of each turn through the ordinary move tools. The moment anything unplanned happens it pauses and hands the unit back with a reason. It never attacks, declares war or ends the turn on its own.

  • Assignments and a notebook that survive amnesia. notebook(action="assign") gives a unit or city a role, a target and a done-when, and every briefing re-checks it against what the seat can see now. remember / recall keep the plan, threats and promises beside the game. Both outlive a context reset, so a model that forgets everything mid-game picks its own plan back up.

  • It looks before it leaps. The tactical view (state(what="unit", id, view="tactical")) is a unit's six neighbours with what a move would do there (attack, open with what the step costs and whose zone of control ends it, refused with why, enemy). state(what="compare") lays out a few production, research or worker options side by side with costs, turns and effects (a building already in the queue reads as queued, not refused). Neither picks for the agent.

  • Batches and safe retries. batch(actions=[...]) runs a list of orders and stops at the first refusal. An action_id makes a retried call a replay, never a second move.

  • A refusal says what to do instead. A mission name the engine does not have comes back with the nearest real ones and the unit's legal orders; a sleep where the unit can fortify names the equivalent; an end refused for units with movement left carries the exact skip orders for the next finish; a greyed button, a refused purchase, peace or ranged attack carries the engine's own sentence; a melee attack the engine drops without a word is named as a refused attack with the first closed gate the harness can read, not as an occupied plot. A refused building names the rule from the game's own table (a Windmill off flat land, Big Ben without Commerce, a resource the city lacks); a refused ideology tenet names its level gate, an unknown policy name comes back with the real names behind the shown ones, and a Great Prophet that has spread its religion is told that is what keeps it from a Holy Site. Every refusal is a next step, never a dead end.

  • A Great Person's withheld missions are explained. state(what="unit", id, view="actions").missing says why the engine greys a mission out: the Great Work slot census (filled / total, the buildings that carry one), the nearest plot where a concert tour or a trade mission plays, no moves left. A concert tour's reply says the tourism and who received it; a gift row names the city-state whose plot the unit stands on.

  • A notebook belongs to one game. Its file is keyed by the game's identity (the seat's starting plot and every slot's civilization, fixed when the map is made), so a new game that founds the same capital on the same turn never opens an older game's notes, assignments or standing orders; a notebook from before the key existed is taken over once, marked adopted, its orders paused until the seat looks at them.

  • A human seat is asked on the trade table. In hotseat, a Declaration of Friendship to another human is a deal that seat answers on its turn, war is the stock War button, denouncing is refused with why; nothing a human seat is sent is answered by its dormant AI. A proposal shows only on the seat it is sent to, and the answer to one echoes its items.

  • It knows when it has won, and when it has lost. The last spaceship part's reply says game_over and victory: "science", the spaceship screen reads complete, and the game_over gate names session(action="main_menu"); the engine's own end-turn blocker freezes at that moment and is not the read to trust. A capital taken ends with defeat: the last city, where it stood and who took it.

  • A turn is one call when nothing needs a look. turn(action="finish", actions=[...]) runs the closing orders and ends the turn together, ending nothing if an order is refused; a move's result carries revealed, what the move showed, so "move the scout and see" needs no second read.

  • It says what a human sees on the leader screen and on the map. Every war the seat is in comes with the Negotiate Peace gate (wars, the briefing's at_war rows: peace can be asked now, or why not) -- an agent sat twenty turns in a siege that one white-peace offer ended. A unit the engine moved while the others played is a unit_displaced event with where it went and what shares the tile it left. A frontier plot a foreign unit or city stands on says occupied, names the city, and stays in the reply however long the frontier is. A neighbour's Great General taking land is named with the plots it took, not just the headline. A turn no wait of ours opened still gets its turn-start work before its first order, and a turn that began while a leader screen was being answered is handed back unended with what it needs, never ended blind.

  • You can watch the seats think. python -m harness.spectator --ledger logs/calls.jsonl serves a page (a canvas map that stays at the frame on Huge maps, d3 for the zoom and the panels) where every plot a tool reply handed a model pulses in that seat's colour over a greyscale map, acts stay painted until the turn ends, the last seat's footprint ghosts through the next, and an Observer / Seat N switch cuts between the unfogged world and what one seat's replies contained. A seat view draws only what that seat's screen shows: never-seen plots black, fogged plots dimmed, other players' units only where visible. Read-only, records and replays with a scrubber (play/pause, speed, a slider whose ticks are the turn boundaries), and reads the game only when the world moves (docs/VISUALIZATION.md).

  • You can see what a turn cost the model. An MCP server never sees a token count; the client does, and writes it to its own files. scripts/ledger_report.py logs/calls.jsonl --usage <the client's records> joins those records (a Claude Code transcript, a Codex rollout, the seat loop's Grok stream) to the call ledger and prints, per game turn, the requests, the new and the cached input, the output, the thinking and a list-price estimate; every session's per-turn sums equal the client's own total. scripts/usage_measure.py says what grows: on a late-game board a late turn of a 20-turn session reads three times an early one, two sessions of 10 turns read 0.62 to 0.88 of one of 20, and the reply that ends a turn is 21 to 55% of what the replies add to the context (docs/measurements/2026-11-14-what-grows/). The spectator page shows the same counts per seat while a game runs, this turn and this game (python -m harness.spectator --usage <the client's records>); a cycle still running is counted request by request. Operator's side only: no seat ever reads a count (docs/TOKEN_ACCOUNTING.md).

  • The largest replies say a thing once. That measurement pointed at four replies, and each was cut where it repeated itself or spoke to nobody: the reply that ends a turn leaves out the status flags that are false, the note it put on every automated unit and the events that are the runtime's own bookkeeping; the World Congress read is a summary by default (detail="full" is the screen's wording); a leader's screen says how that leader stands with you without the flags that are false. Read back on Claude Code, Codex and Grok on one board: the reply that ends a turn is 12 to 32% smaller than in 2.15.0, the league read 31 to 42%, and no client made an extra call for what is gone. A whole session did not get measurably cheaper: its reading is its requests times its context, and the number of requests varies more than a reply saves.

  • briefing measured live. Over four hotseat turns played both ways (docs/NOTES.md): 4.25 calls a turn against 8 with the separate reads, no refused order against two, about the same bytes.

  • The tactical view never guesses. Adjacency and map wrap come from the engine. One step's cost and the zone of control that ends it follow the engine's movement rules on the plots as seen, and reach walks that rule to every plot this turn's move can end on (2.14.0; 34 live orders arrived as listed); no turns-to-reach beyond this turn, which the engine cannot give safely. Nothing fogged is called safe.

  • A compare keeps a tile's own gain apart from the empire's (nothing until a city works it) and never calls a trade destination safe under fog.

  • A trade unit's second gate. Past the slot count the engine asks whether some city of yours can start a new route of that kind now; when it cannot, the briefing, compare and a refused production order quote the engine's own sentence ("no available sea trade routes") instead of offering to build or buy.

  • What an assignment review can report. on_track, condition_met, or needs_review with the observation behind it: a unit gone (an upgrade on its last plot is named; a unit our own order used up -- a Great Work written, a religion's last spread, a gift to a city-state -- is unit_spent with the reason, a plundered caravan says plundered; our own upgrade order carries the assignments and the open order onto the new id by itself), a reused id never followed, a city lost, a settle site now too close to a city, or a target out of sight kept as last seen, never assumed gone. notebook(action="amend") and notebook(action="close") do the rest.

  • What pauses a standing order. A hostile in sight, damage, an enemy on the destination, a refused step, no progress, or a direct command to that unit (never the command that created it).

  • What wakes skip_quiet_turns. Combat of our own, a leader message that asks something, an empty city, a happiness drop out of the comfortable band (or five at once), a deal in its last turns (once), a new strategic-resource shortfall, or a barbarian camp or hostile unit beside a unit on its way somewhere. A caller's wake_on words always wake; a fight between other players, their alliances, denouncements and quests, and a splash the harness swept never do.

  • After a context reset, a briefing carries the latest notes again and every assignment is re-checked.

What a game looks like

The agent plays through tool calls; you watch the game window. On its turn it checks what changed, inspects its cities and units, chooses actions the game permits, then ends the turn. In hotseat the window passes back to you and you take your turn as usual. Nine tools -- turn, state, order, question, answer, notebook, batch, session, reference -- and the loop is four lines:

turn(action="finish")                      -> the new turn: status, digest, notes; read `gate` first
question() -> answer(choice)               -> whatever the game asks, from its own `answers` rows
state(what="units", view="actions") ...  -> every reply that lists what can be done carries the order to copy
order(to="unit", id, action="move", args)  -> then notebook(action="remember") and turn(action="finish") again

When something blocks the turn, the status names it and hands back the rows that clear it. Refused orders explain why, the shape they should have had, and an example. The 145 one-tool-per-button names of the 1.x releases still answer, each reply naming the call to copy instead.

Turn 269, trimmed for width. It predates runtime v217; a status now also names the seat, carries the gate to clear first, happiness and any alerts, and lists automated or already-moving units under todo.ongoing:

{"turn": 269, "my_turn": true, "mode": "single", "blocking_name": "NO_ENDTURN_BLOCKING_TYPE",
 "todo": {"promotions": [], "research_unset": false, "cities": [], "units": []},
 "pending_popups": [], "game_over": false, "notifications": {"live": 2, "held": 99}}

What to expect

  • It is slow and it costs tokens. A developed empire means dozens of tool calls per turn. A game to victory is a long project; an evening is a few dozen turns. Quiet-turn skipping and standing orders are what keep that bill down.

  • Leave the window alone during the agent's turn. It drives the game's own screens the way a mouse would, and a stray click from you lands in its turn.

  • The game crashes sometimes. The Linux port does, with or without the harness. Quick saves every turn in solo (session(action="save") is one call elsewhere), the game's own autosaves and session(action="load_latest") make it a pause, not a loss. A supervisor can relaunch the game and rejoin a LAN game on its own.

  • You can read its mind. Every note the agent writes to itself and every assignment it gives a unit is a plain JSON file under ~/.local/share/civ5-harness/notes/, per game and seat. You can see what it planned before you find out whether it worked.

Ways to play

Mode

Who is where

Start it with

Solo

LLM in seat 0 versus the game's AI

harness.cli start-single --civ CIVILIZATION_ROME

Hotseat

You and the LLM(s) at one machine, turn by turn

harness.cli host-hotseat --humans 0 1 --nick 1=Claude

LAN

Each LLM runs its own game instance and joins like any player

scripts/launch_llm_client.sh, then harness.cli join-lan <host>

All three share the same tools. Several agents in one game each get their own MCP server, one per seat; every status names the seat it is playing, and session(action="seat") moves a server to another human seat without a restart. The agent cannot take over one of the built-in AI seats; the game does not expose that. To move a two-seat hotseat game along between sessions, scripts/hotseat_rounds.py alternates the seats and stops at the first turn that needs a decision.

The install guide walks through each mode. Saved states in saves/ reproduce late-game diplomacy, peace terms, Venice puppets and a combat lab if you want to drop a model into something interesting on turn one.

How it works

flowchart LR
    A["LLM client"] -->|stdio| B["MCP server<br/>nine tools"]
    B -->|Unix socket| C["tunerd<br/>one tuner connection"]
    C -->|TCP 4318| D["Civilization V<br/>preload shim + FireTuner"]

Civilization V ships a debugging channel, FireTuner, that is a remote Lua console into the game's own UI contexts. The harness keeps that channel open in multiplayer with a small preload shim, injects a Lua runtime that ports the stock UI's own logic (combat previews, trade legality, hovers), and drives every screen the way a mouse would. Because the network only carries player commands in a lockstep simulation, a real game instance is the only faithful client; that is why the LLM gets one. docs/ARCHITECTURE.md has the long version.

Honest limits

  • Linux only, native Steam build only, Brave New World only. No Windows, no macOS, no Proton.

  • With the tuner enabled the game listens on every interface by default. The install guide binds it to loopback; do not skip that on a shared network.

  • A blocker the engine raises without opening its chooser (the Maya Long Count, a stolen tech, a free Great Person, a dig) is asked and answered through question() / answer() since 2.2.0; anything still outside the nine tools makes the agent stop and ask you. docs/GAPS.md is the running audit.

  • The engine's movement-cost preview and path overlays crash the game when read, so one step's cost is computed from the engine's movement rules instead, and nothing beyond one step (turns to reach) is estimated.

  • Token counts are the client's own, read from the files it writes; a client that writes none gets none, and nothing is counted with a tokenizer. The dollar column is a list-price estimate from a dated table (harness/usage/prices.json), not what a subscription was charged; a model the table does not list prints tokens and no dollars.

  • No hosted CI by choice; the suite runs locally before every push.

Documentation

Development

sudo apt install lua5.4 liblua5.4-0   # luac for the runtime lint, liblua for the Lua tests
uv sync --group dev
scripts/check.sh            # 2276 tests: the shipped Lua under lupa, the Python layer against fake bridges; no game needed

CIV5_CALL_LOG=/path/calls.jsonl in the server's environment writes one line per tool call (bytes, trips, seconds, refusals, and client: CIV5_CLIENT when set, else the MCP client's name/version, so two models on one seat stay apart); scripts/ledger_report.py [--seat N] [--client codex] turns it into a per-turn table with a per-client footer, and --usage <dir> adds what each turn cost the model in tokens, read from the clients' own usage records (docs/TOKEN_ACCOUNTING.md); scripts/usage_measure.py <ledger> --usage <dir> writes one document per client and model on what grows over a session. Off by default; scripts/llm_seat_loop.py sets both for the agent it runs.

python -m harness.spectator --ledger /path/calls.jsonl serves a live, read-only page (web/viz/, d3) of the game and the seats' attention: the whole map in greyscale, every plot a tool reply handed a model pulsing once in the seat's colour, acts staying painted for the turn, a ghost of the previous seat, a heatmap, and an Observer / Seat N switch. --record / --replay work without a game. --usage <dir> adds tokens per seat, this turn and this game, counted by the report's own code from the same usage records (the operator's view; no seat reads the page). See docs/VISUALIZATION.md.

The Lua runtime is one file per game domain under harness/lua/runtime/ (its README.md says which file owns what and how to add one); harness/runtime_source.py lists the load order and assembles what is injected. It carries its own version counter (RUNTIME_VERSION in bootstrap.lua), bumped on every change to any of those files, because a running game keeps the old runtime until a newer number arrives; the source digest re-injects a changed runtime even when the bump was forgotten. Commit subjects carry it as runtime vNNN.

Issues, merge requests and milestones live on GitLab, the only write-enabled repository: https://gitlab.com/Tyler-Meador/civ-v-linux-mcp. CONTRIBUTING.md has the rules a change has to keep; the issue and merge request templates are under .gitlab/.

Authorship

This project was written by Claude, Anthropic's AI model, working in Claude Code under the direction of Ty Meador, who owns the game, ran every live session and decided what the harness should and should not do. Claude Fable 5.1 wrote most of it, with Claude Opus 5, Claude Sonnet 5 and Claude Opus 5.5 on other commits; the Co-Authored-By trailers in the git history say which model committed what. The fixes of 2.11.0 to 2.13.0 (2026-10-31) were drafted by Claude Opus 5.5 subagents and reviewed, corrected and committed by Claude Fable 5.1, so their trailers name the reviewer. That covers the reverse engineering of the FireTuner protocol and the game binary, the shim, the Lua runtime, the MCP server, the tests, the documentation and this README. Ty's contribution is the design brief, the live verification against the running game, the judgement calls recorded in docs/GAPS.md, and the standard that the seat may see only what a human sees.

License

MIT. Sid Meier's Civilization V, its assets and the stock UI Lua quoted in docs/ belong to Firaxis Games and Take-Two Interactive and are not covered by this licence; you need your own copy of the game.

Available Tools

9 tools
answerA

Answer what the game is asking: settle the one popup, leader screen or deal that question() shows, and nothing else. choice is the answers row's choice: an item's name (a TECH_, UNIT_, BUILDING_, POLICY_, PROMOTION_, BELIEF_ or goody name; liberate / annex / puppet / raze; a dig's choice number), a popup's or leader screen's button number, or on a leader screen accept / refuse (the deal on the table), leave (close it), accept_friendship. args carries the rest the popup needs (a promotion's unit_id, a dig's x and y, a stolen tech's player_id, a religion's beliefs and city). Dispatches on the gate: question() shows which question is up and hands back the rows to copy. It acts on the game as the popup's own button would -- an accepted deal, a razed city, a chosen tech cannot be taken back; a choice the popup does not offer, or an answer when nothing is asked, is refused with {ok: false, err, choices} and nothing happens. action_id: a retry with the same token replays the stored reply instead of answering twice. Not this tool: a decision with no popup up (research, production, a policy) is order; what a choice means is reference. Returns {ok, ...} with the popup's outcome and the gate that follows (still_pending and next when another leader waits; next also names the call that resumes the AIs' round when they are still moving), or the refusal.

Inputs: choice (required), args, action_id.

batchA
Destructive

Many orders in one call, in order, with the turn left open: actions is a list of {tool, args} rows -- the rows every read hands back (actions, call): order, answer and notebook rows alike -- each run as that call and answered under results. The first refusal stops the rest by default, since the state the later rows were reasoned about is no longer certain (stop_on_refusal=false runs them all); the rows not run come back under skipped. force=true takes over this turn's claim when another client of my seat holds it (otherwise the first row is refused with turn_claim). Not inside: the turn's wait / finish / end, a load, the seat, the exit, a batch. Each row changes the game as its order would and cannot be undone; a row's own action_id replays. Use it for several orders already decided; use order for one, and turn(action='finish', actions=[...]) when the same rows should end the turn afterwards. Returns {ok, done, results: [one reply per row], skipped?, hint?}: ok is false when any row was refused.

Inputs: actions (required), force, stop_on_refusal.

notebookA
Destructive

My memory, kept beside the game per seat and surviving a context reset or a restart: free-text notes, assignments (what a unit or city is for) and standing orders (a unit's plan the harness carries out over the coming turns). Use it for what future-me must know; the game itself holds no plan. Notes and assignments move nothing; a standing order does: give_order and resume_order run its first steps NOW (a move spends this turn's movement, a build starts) and carry on at the start of each of my turns. action, with its args -- remember {text, tag?, replace_id?}: write a note (returns it with its id; new notes ride on the next briefing / finish); recall {tag?, limit?}: list notes; forget {note_id}: delete a note for good. assign {role, purpose, unit_ids? / city_ids?}: give units or cities a purpose that shows in their state rows; assignments: list them; amend {assignment_id, purpose? / role?}; close {assignment_id}. give_order {unit_id, steps, interrupt?, purpose?, replace_id?}: steps is a list of at most 6, in order -- {kind: 'move', x, y}, {kind: 'build', build: 'FARM'}, {kind: 'heal', hp: 80}, {kind: 'hold', mission: 'fortify' | 'sleep' | 'alert'} (last) -- each step runs at most once per turn and the next starts as soon as one completes (a move that arrives starts its build the same turn); it never attacks, declares war or ends the turn, and pauses with a reason when a hostile comes near, the unit is hurt or a step is refused; one open order per unit (replace_id replaces it). orders: the plans with their status; cancel_order / resume_order {order_id}. Not this tool: a one-off move is order(to='unit'); a popup is answer; the board is state. Usable while it is not my turn. Returns {ok: true, ...} with the row written or the rows listed, or {ok: false, err, accepts, example} (an unknown action or a missing argument; nothing is written). A note written on a turn later than the one on screen (a save was loaded) comes back marked later_timeline.

Inputs: action (required), args, action_id.

orderA
Destructive

An order to something I own: to the target kind, id which one, action the verb, args its arguments by name. It changes the game, and most of it cannot be undone (a disbanded unit, a declared war, a purchase). Not a read: state shows the board and hands back the rows, question / answer settle what a popup asks, turn(action='finish') ends the turn. to: actions -- unit: move, mission, promote, upgrade, disband, trade_route, plunder city: produce, purchase, dequeue, sell_building, buy_plot, specialist, work_plot, governor, status, strike, great_work empire: research, policy, ideology, switch_ideology, pantheon, religion, enhance_religion, reformation, faith_purchase civ: propose_deal, negotiate, peace, demand, friendship, declare_war, denounce, event city_state: gift_gold, gift_unit, gift_improvement (or a city-state screen button by its name: pledge, revoke_pledge, bully_gold, bully_unit, declare_war, make_peace) league: propose, repeal, vote spy: move, coup How: copy the rows a read hands back (actions, call) as they are; id is a unit, city, civ, city-state or spy id (empire and league take none); args holds what the row holds (a plot's x and y, an item's name, a deal's items, an amount). An order runs whole or not at all. action_id: a token of mine; a retry with the same one replays the stored reply instead of acting twice. Returns {ok: true, ...} with what changed (a unit's new plot and moves left, a city's queue, a deal's answer), or {ok: false, err, choices / accepts, example}: the verb is not one of the target's (choices lists them), an argument is missing (accepts names the shape, example a complete call), or the game refuses it (err names the rule, e.g. the unit has no moves, the city is not mine, another client of my seat holds this turn's claim). Nothing is acted on when ok is false.

Inputs: to (required), action (required), id, args, +1 more.

questionA
Read-onlyIdempotent

What the game is asking right now, so answer() can settle it. Read-only; takes nothing. Use it whenever a reply's gate is not null or turn(action='status') says discussion_pending; use state for the board and reference for the rules. It shows one of: a decision popup (a tech, a production item, a policy, a promotion, a belief, a ruins reward, a captured city's fate, a Great Person's use, a dig), the leader on screen with their words and the deal on the table, another human seat's proposal waiting on the trade table, or nothing pending. Returns the popup's own read -- the question in words plus answers rows, each a complete answer() call to copy (the item's name or button number, and the args the popup needs) -- with gate on the reply; a leader screen returns the leader, the speech, the buttons and the deal's items; when nothing is asked, {pending: false, decisions: [...]} lists the turn's open decisions instead. Several leaders can queue at a turn start: answering one returns the next under next.

Inputs: no named parameters.

referenceA
Read-onlyIdempotent

What things do, in two books. book="rules" (the default): the rule book of this very game, read once from its database (mods and DLC included) and cached -- every unit, building, wonder, project, process, promotion, social policy and ideology tenet, technology, belief, resource, terrain, feature, improvement, specialist and unit action, with its numbers. book="playbook": how to play through these tools (the turn loop, batches, blockers, diplomacy, the house rules, the action table). Read-only, usable while it is not my turn. Use it to learn what a choice means; use state for the live numbers (its chooser rows carry only the enum, the name and live numbers, never this text) and question for what a popup asks. section: one part per call. Rules sections: terrain, resources, improvements, units, buildings, projects, processes, promotions, policies, techs, beliefs, specialists, actions; left out, the whole book (tens of thousands of tokens: read it once at the start of a game if you can hold it, else the section a choice needs). Playbook topics: start, turn_loop, quiet_turns, batches, verify, blockers, diplomacy, rules, first_turn, orders (the action table), all; left out, the index of topics with the start section; city_state(s), deals, trade and league answer with diplomacy. Returns Markdown. An unknown rules section is refused as {ok: false, err, sections}; an unknown playbook topic answers with the topic list. Also served as the MCP resources civ5://reference and civ5://reference/{section}.

Inputs: book, find, section.

sessionA
Destructive

Saves, loads and the seat of this game session: keep a state before a risky turn, open another save, leave a finished game, say which hotseat seat this server plays. It never plays a turn; use turn for that. Side effects: a load replaces the game on screen for everyone on it, and a hotseat game loads paused until turn(action='wait'); the notebook is per game and seat, so it follows a save of the same game and is not touched by a load. action -- save: a quick save of the game on screen into the one quick-save slot (each finished turn's autosave overwrites that slot, so copy a save you want kept); load {filename}: open that save (a name as the game lists it, without the single/ prefix; the refusal lists the legal ones), replacing the unsaved state, so save first; load_latest: open the newest save, whichever game it is; main_menu: leave the game for the main menu (the game_over gate's way out); seat {seat}: which human seat (0 or 1) this server plays in a hotseat game -- a server pinned with --seat refuses a change onto the other human seat, and never move onto the seat on screen. Returns {ok: true, saved / loaded, turn, seat, screen} or {ok: false, err} naming why (an unknown filename with the list, a seat that is someone else's, a game operation already running). In a game with other people ask the human before loading.

Inputs: action (required), args.

stateA
Read-onlyIdempotent

Read the game as my seat sees it: the empire, units, cities, civs and city-states, the map, the league, spies, the screens, and the choices open to me (research, policies, beliefs, trade routes). Read-only, no side effect, usable while it is not my turn. Use it before an order (its rows ARE the orders) and after one to verify; use reference for what things do, question for what a popup asks, turn for the turn's own state. How to call: what names the object, view the part of it (left out: the default view), id the one object. Objects and their views: empire: default, demographics notifications: default screen: tech_tree, religion, culture, culture_works, great_people, domination, spaceship, wonders, intrigue, trade_routes, deals league: default spies: default units: default, actions unit: actions, targets, tactical, home cities: default city: screen, production, strikes map: window, grid, index, frontier civs: default civ: summary, war, trade_catalog city_state: summary, full options: research, policies, beliefs, trade_routes, spy_cities compare: default Selectors, by view: x, y, radius for map/window and map/grid; radius alone for unit/tactical (centred on the unit); id for map/frontier (the explorer) and options/trade_routes (the caravan); mission for unit/targets; kind for options/beliefs and compare; detail ('summary' / 'full') where a view has both; since and limit for lists; item and yield_type for city/production to price one item; args for the rest by the old call's name. Returns: the object as JSON (a list for units, cities, civs; one dict otherwise), with actions or call rows wherever something can be done: each row is a complete order() call to copy. A wrong what or view is refused with choices, and for an object that takes an id, a hint saying where the ids come from (unit ids from units, city-state ids from the minor: true rows of civs). Replies are the 1.x reads' replies unchanged.

Inputs: what (required), x, y, id, +10 more.

turnA

The clock: my seat's turn loop, one action per call. Actions: status -- whose turn it is, the gate (answer it before anything else), what blocks the end. Read-only. briefing -- my turn in one read: decisions with their actions rows, changes, board, threats, notes. finish -- run actions as a batch, end my turn, wait until it is mine again, return the new turn. end -- end the turn only, no wait. wait -- block until it is my turn again. Arguments by action: since, limit and detail belong to briefing (since='turn' after a context reset; detail='digest' is the event digest alone); actions and notes to finish; timeout_seconds to finish and wait; args (skip_quiet_turns, wake_on, autosave, force, briefing) to finish and end. Side effects: finish and end hand the game to the other players, and nothing they do can be undone -- order and answer everything first; a session save is the way back. finish autosaves to the quick-save slot. A refused row in actions stops the batch and the turn is not ended. Safe to repeat: when it is not my turn, finish only waits. A second client of my seat is refused with turn_claim (args.force takes the turn over). Timeouts: past timeout_seconds (600 by default) the reply says my_turn false, timed_out true and next is the call that resumes the wait; progress notifications go out every 5 s meanwhile, so one long call beats several short ones. Not this tool: question / answer for what the game asks (nothing ends while gate is set); state for the board; batch for orders that leave the turn open; session for saves. Returns: turn, my_turn, gate, todo / decisions, digest (what happened meanwhile), notes; game_over with victory / defeat at the end. A refusal is {ok: false, err, end_turn: {todo, skip_actions}}: what blocks the turn and the rows that clear it.

Inputs: action (required), args, limit, notes, +4 more.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.6/5.0

    Scored across 9 tools

    Disambiguation5/5

    Each tool owns a distinct slice of the loop: order (single mutation), answer (popup settlement), state/reference/question (three different reads), turn (clock), batch (multi-order), notebook (memory), session (saves/seat). The descriptions go further than most servers by including explicit "Not this tool" sections (e.g. order vs batch vs turn(finish), answer vs order) so the nearest-neighbour confusions are pre-empted.

    Naming Consistency4/5

    All nine names are single lowercase words with no prefixes, suffixes, casing or separator mixing (order, answer, batch, session...), which is highly predictable. The only weakness is that they are bare nouns rather than verb_noun, so a few (state, notebook, reference) don't self-describe their action as immediately as a verb would.

    Tool Count5/5

    Nine tools for a game harness spanning mutation, reads, popup resolution, turn management, batching, persistence and session control is tight and every tool has a clear distinct job. Complexity is pushed into per-tool `action`/`what` sub-selectors rather than into more top-level tools, which is the right trade-off here.

    Completeness4/5

    The lifecycle is essentially covered: read board/rules/prompts, issue orders singly or batched, settle every popup, advance/wait the turn, keep notes and standing orders, save/load/seat. The one plausible gap is that session only saves, loads and reassigns seats — there is no operation for starting a new game, which a full session lifecycle would normally include.

    Maintenance

    ActivityActive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      D
      maintenance
      An MCP server that lets LLM agents play full games of Civilization VI. It connects to a running game and provides tools for unit movement, city management, diplomacy, and more, all through the game's rule-enforcing APIs.
      76
      241
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables any LLM to play Freeciv as a player through MCP tools and resources, with a graphical observer view.
      MIT
    • A
      license
      A
      quality
      B
      maintenance
      Enables LLM agents to play full games of Civilization VI by connecting to running game instances via the FireTuner debug interface, with 76 tools for units, cities, diplomacy, research, and turn management, plus narration of game state through rule-enforcing APIs without requiring vision models.
      80
      MIT