Local MCP server for the Slay the Spire 2 bridge mod, providing tools to interact with the game state, perform actions, and manage automation.
STS2 MCP server has well-structured tool definitions with consistent naming (verb_noun pattern: sts2_get_*, sts2_perform_*, sts2_resolve_*) and comprehensive descriptions (avg 180 chars, within 10-1024 baseline). All 13 tools have input schemas with proper JSON Schema types. However, parameter descriptions are sparse or missing entirely, most parameters lack explanatory text beyond the name. Output schemas are not documented. Error handling guidance is absent. The server is domain-specific (Slay the Spire 2 game automation) with clear intent, but lacks LLM-optimization patterns like recovery guidance, actionable error messages, and per-parameter descriptions that would help agents self-correct.
Convenience wrapper around sts2_perform_action for the combat end_turn action. Set return_state_after=true to include a compact post-action state summary in the tool response. Before ending: check for unplayed 0-cost cards and unused potions.
Execute a mixed combat sequence in one tool call. Supports play_card, use_potion, and end_turn actions, automatically rematching later play_card/use_potion steps after reindexing or draw changes. If end_turn is included anywhere in action_ids, it is always executed last. Set return_state_after=true to include a compact post-sequence combat summary in the tool response.
Read the Slay the Spire 2 bridge session file and verify the current /health endpoint.
Fetch the current player's full master deck as a low-frequency strategic view, including grouped counts and per-card summaries for shop, upgrade, and boss-path planning.
Build a pruned future-only map route forest from the currently travelable frontier points, excluding the current node, past rows, and unreachable branches. Defaults to summary mode; use detail=full to include the full reachable node table. Strategy tip: prefer routes with more shops and campfires; elite fights have very low reward-to-risk ratio — avoid or minimize them.
Parameter descriptions missing or minimal. Most parameters (action_id, expected_state_version, strict, wait_after_ms, return_state_after, option_index, etc.) lack explanatory text. LLMs cannot infer parameter meaning from names alone, 'strict' could mean strict validation, strict ordering, or strict error handling.
Output schemas not documented. Tools return complex game state objects (screen, combat state, map state, rewards, legal actions) but no schema is provided. LLMs cannot plan downstream calls or extract required fields without knowing the response structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 67 | 2026-07-28+ | v2 |
Fetch the current visible bridge state, including screen, combat state, map state, rewards, and legal actions.
Fetch the currently legal bridge actions without repeating the full state payload.
Execute one currently legal bridge action by action_id. Avoid parallel combat play-card calls; use sts2_play_card_sequence or sts2_execute_combat_sequence for consecutive combat actions. Set return_state_after=true to include a compact post-action state summary in the tool response.
Pick a visible option by index from the current screen, then optionally auto-advance if the screen closes.
Execute multiple currently planned play_card actions in one tool call. Later steps are automatically rematched against the current hand and legal targets after each card resolves, so the caller does not need to manually rewrite hand indices after reordering or draws. Do not parallelize consecutive combat card plays with sts2_perform_action. Set return_state_after=true to include a compact post-sequence combat summary in the tool response.
Resolve the currently visible card-selection flow in one tool call: optionally choose one or more current option indices, then confirm/cancel/skip/close without one MCP request per selected card.
Resolve a campfire in one tool call: choose rest/smith, optionally choose a smith upgrade target, then auto-advance back to the map when possible. Strategy tip: rest if HP < 50%; consider smithing if HP > 70% and you have a key card to upgrade.
Inspect the current room-end rewards, optionally choose one card reward by index, claim safe rewards, and auto-advance through reward/proceed cleanup when possible. Strategy tip: consider your current deck archetype before picking card rewards; choose cards that synergize with your build.
No error handling guidance. Tools lack recovery hints, error categorization (retryable vs fatal), or actionable error messages. If a tool fails, the LLM has no guidance on what to do next.
Ambiguous parameter semantics. 'strict' mode is mentioned but not defined, does it validate state version, enforce action legality, or both? 'wait_after_ms' is capped at 5000 but no rationale is given. LLMs cannot reason about when to use these parameters.
No confirmation or dry-run pattern for destructive operations. Tools like sts2_perform_action and sts2_resolve_room_rewards modify game state irreversibly but lack a dry-run or confirmation step. Agents cannot preview consequences before committing.