GardenArena MCP server

• Registry nameio.gardenarena/gardenscriptverified in the MCP registry

The registry name sits under gardenarena.io. The registry only lets a publisher use that namespace after proving control of the domain through DNS or HTTP.

• Stateanswerschecked 2026-10-26
• Tools30
• Server calls itselfGardenArena GardenScript
• Server version1.28.1
• Protocol version2025-06-18

On 2026-10-26 the server answered initialize and tools/list without a sign-in. First answer in 213 ms.

• Its tools do every use case we checked: run a farm simulation under a fixed rule set; compare two strategies in a head-to-head duel; review replay traces to see why a policy lost.

Said by rules from the measurements above. The likely first call is a model's pick from the tool list; read its description before relying on it.

• INITE MCP score77 of 100answers 100 · does the job 100 · definitions 87 · context 90 · connects 100 · lowered by the warning signs below
• What an agent can do with it3 of 3 use cases
• Answered100% of days8 of 8 daily checks
• Median first answer532 ms
• Space in an agent's context≈ 4,418 tokensthe definitions of all its tools (30), which an agent carries on every turn
• Tool definitionsB · 87 of 100
• Typed results30 of 30 toolsdeclared outputSchema: results come typed, read once

• Parameters without a description: in 24 of 30

A program checks the server once a day: initialize and tools/list, no model. Tokens are the definitions' length at about four characters a token. The grade averages six checks of every tool's name, description, schema and annotations; it says how well the tools are described, not whether they work. The score weighs what was measured and nothing else: answering (25%), doing the product's job (25%), definitions (20%), context (10%), use (10%) and how readily a client connects (10%). A part that was not measured is left out, and a hidden instruction caps the score at 40. The state of vendors' MCP servers: method and numbers →

• validate_gardenscript

  Validate and compile an untrusted GardenScript source with the local Rust compiler. This performs no LLM call and stores nothing.

• get_bamboo_scene

  Read one day (1-365, default UTC day of year) of the published Rust bamboo replay used by the 3D scene. Includes sensors, entrant and SHA-256 provenance. Synthetic exhibition only; not an observation, strategy evaluation or hydraulic…

• bamboo_challenge_get

  Read the fixed 365-day bamboo training mission, exact Dry Patience reference, scenario hash, budgets and quotas. No run or publication. Read gardenarena://gardenscript/v0.5 for the full language.

• coop_workshop_get

  Read the playable chicken workshop mission: breeds, breeding/incubation, production balance, non-lethal predators, closed JSON commands, exact scenario and quotas. No simulation or database write. Distinct from GardenScript/official…

• coop_workshop_run

  Execute one bounded Rust chicken-workshop journal after human agreement. journal_json is the complete closed JSON journal (64 KiB), NOT GardenScript or arbitrary code. Returns reproducible state, event explanations and an export for…

• farm_workshop_get

  Read the playable whole-farm mission, shared resources, closed JSON actions and farmer task contract. Educational simulation, no run or official write.

• farm_workshop_run

  Execute one bounded whole-farm Rust journal after human agreement, within the agreed call budget. journal_json is complete closed JSON, max 64 KiB, not GardenScript. Returns game state, evidence-linked farmer tasks and an export for…

• farm_next_event

  Advance a farm journal to the next detected important moment, checked daily for at most max_days (1–21). Use the full journal_json and care_json containing ration, access, ventilation, incubator. Consumes one agreed calculation call…

• farm_autopilot_run

  Run a bounded pantry-oriented farmer tour after user agreement: 1–7 simulated days, 0–1000 game credits, at most 64 added ordinary commands. Deterministic Rust rules, not a live LLM. Gathers food, handles care and irrigation, cuts mature…

• farm_campaign_get

  Read chapter 2 stewardship-v1: exact teaching farm J95, three management choices, ten-day routine, success criteria and complete original journal. No calculation or publication; not the player's private farm. JSON plan, NOT GardenScript.

• farm_campaign_run

  Try the fixed J95 to J105 farm chapter after agreeing a finite call budget. choices_json is a CLOSED object: mission_id stewardship-v1, choices {harvest: pantry/kitchen/market, egg_reserve: 0/2/6, rotate: boolean}. At most ten days, 100…

• farm_policy_get

  Read GardenScript 0.6 farm.v1: six decision channels, 29 current-state sensors, exact intention expansion, three application-authored presets and shared limits. No run, persistent write or LLM call. Read before writing a farm policy.

• farm_policy_validate

  Validate a closed GardenScript 0.6 farm.v1 source, maximum 16 KiB. Local Rust only; line diagnostics and exact source/canonical hashes. Shared workshop quota; no run, persistence or publication.

• farm_policy_run

  Run a GardenScript 0.6 farm.v1 policy from a COMPLETE v4/v5 journal after user agreement, within a finite call budget. Six decisions from each day's current snapshot become ordinary costed commands. No future weather archive access or…

• farm_compare_get

  Read the bounded two-policy farm comparison contract and its fixed week-after-v1 teaching mission (exact J96 journal, two sources, seven days, 540 credits). V4 only, same start and days/budget per branch, no global winner or publication…

• farm_compare_run

  Compare two GardenScript farm.v1 policies from the SAME complete v4 journal after agreeing a finite call budget. Two trials, 1–7 days and identical 0–1000 game-credit ceilings. Rust returns both branches, trace witnesses of the first work…

• farm_weather_search

  Find French metropolitan communes by name or postcode for opt-in farm weather. Public administrative data, cached and bounded. No IP geolocation, precise address, game mutation or LLM call.

• farm_weather_get

  Read cached real MET Norway forecasts and a continuous Météo-France station archive for a chosen French commune INSEE code. Optional start_date and station_id retain an existing game's station/start when refreshing. Returns a frozen…

• bamboo_training_run

  Consume one bounded private training run against Dry Patience after user agreement. GardenScript 0.5 bamboo only, fixed public scenario and budgets. Shared web/MCP quotas; temporary local storage, no official submission or publication, no…

• bamboo_training_explain

  Read a private training comparison and evidence-linked moments. Optional day 1-365 returns both actual Rust decision traces, not a rerun. Requires the private GardenArena run_token, not a provider credential.

• bamboo_training_artifact

  Read the exact reproducible request and expected result hashes for a private training run. Includes strategy sources: do not publish automatically. Full trace JSON downloadable through the documented POST export. No run or official ledger…

• list_garden_arenas

  List GardenScript arenas. This is local, read-only, and performs no LLM call.

• get_garden_arena

  Read one arena's ruleset, scenario, capabilities, limits and status by slug. This performs no LLM call.

• submit_garden_strategy

  Submit externally generated GardenScript to an arena. The source is validated again by the local Rust compiler. Disabled by default; operator-owned identity comes only from server environment, never from a model token.

• get_garden_results

  Read the sealed results for a GardenScript arena. Performs no simulation or LLM call.

• entry_create

  Create a Reality Trial 001 policy entry. Returns an explicit entry_id. Returns a private entry_nonce once; keep it for all mutations. An idempotent retry requires that nonce and never returns it again. Identity assurance can only be…

• strategy_validate

  Compile GardenScript for an existing entry without running the official scenario. Requires its private entry_nonce. Consumes that entry's validation budget.

• strategy_prepare_seal

  Prepare a review and return its confirmation_url for the human organizer. Requires the private entry_nonce. Does not grant authority to seal. The organizer must use their own web session; never operate that page for them.

• strategy_seal

  Read acknowledgment of an already human-confirmed seal. This tool can NEVER create a seal: pending entries return human_confirmation_required. Give the preparation's confirmation_url to the organizer instead. Legacy receipt/handle…

• artifact_verify

  Verify a SHA-256 digest against published contracts, scenarios and sealed entries.