Too many MCP tools

[Gennadiyev]

There are currently 55 MCP tools.

- mcp__sts2__get_game_state
- mcp__sts2__use_potion
- mcp__sts2__proceed_to_map
- mcp__sts2__combat_play_card
- mcp__sts2__combat_end_turn
- mcp__sts2__combat_select_card
- mcp__sts2__combat_confirm_selection
- mcp__sts2__rewards_claim
- mcp__sts2__rewards_pick_card
- mcp__sts2__rewards_skip_card
- mcp__sts2__map_choose_node
- mcp__sts2__rest_choose_option
- mcp__sts2__shop_purchase
- mcp__sts2__event_choose_option
- mcp__sts2__event_advance_dialogue
- mcp__sts2__deck_select_card
- mcp__sts2__deck_confirm_selection
- mcp__sts2__deck_cancel_selection
- mcp__sts2__bundle_select
- mcp__sts2__bundle_confirm_selection
- mcp__sts2__bundle_cancel_selection
- mcp__sts2__relic_select
- mcp__sts2__relic_skip
- mcp__sts2__treasure_claim_relic
- mcp__sts2__crystal_sphere_set_tool
- mcp__sts2__crystal_sphere_click_cell
- mcp__sts2__crystal_sphere_proceed
- mcp__sts2__mp_get_game_state
- mcp__sts2__mp_combat_play_card
- mcp__sts2__mp_combat_end_turn
- mcp__sts2__mp_combat_undo_end_turn
- mcp__sts2__mp_use_potion
- mcp__sts2__mp_map_vote
- mcp__sts2__mp_event_choose_option
- mcp__sts2__mp_event_advance_dialogue
- mcp__sts2__mp_rest_choose_option
- mcp__sts2__mp_shop_purchase
- mcp__sts2__mp_rewards_claim
- mcp__sts2__mp_rewards_pick_card
- mcp__sts2__mp_rewards_skip_card
- mcp__sts2__mp_proceed_to_map
- mcp__sts2__mp_deck_select_card
- mcp__sts2__mp_deck_confirm_selection
- mcp__sts2__mp_deck_cancel_selection
- mcp__sts2__mp_bundle_select
- mcp__sts2__mp_bundle_confirm_selection
- mcp__sts2__mp_bundle_cancel_selection
- mcp__sts2__mp_combat_select_card
- mcp__sts2__mp_combat_confirm_selection
- mcp__sts2__mp_relic_select
- mcp__sts2__mp_relic_skip
- mcp__sts2__mp_treasure_claim_relic
- mcp__sts2__mp_crystal_sphere_set_tool
- mcp__sts2__mp_crystal_sphere_click_cell
- mcp__sts2__mp_crystal_sphere_proceed

I think it is time to reduce the load. The two options are:

1. Use skills instead of dozens of MCP tools.
2. Accept --singleplayer or --multiplayer as a MCP server launch option, effectively halving the number of tools.

[Gennadiyev]

Good news: #37 adds more tools!

[CompleteDotTech]

Would you like assistance consolidating some of these multi click tools into single tool call tools?

[judgeou]

https://vercel.com/blog/we-removed-80-percent-of-our-agents-tools

I believe that reducing the number of MCP tools is valuable.

[Gennadiyev]

https://vercel.com/blog/we-removed-80-percent-of-our-agents-tools

I believe that reducing the number of MCP tools is valuable.

Agreed.

As per our README, we are now recommending skill as a drop-in replacement for our MCP server. Hopefully, #44 will be the PR to address this.

[Gennadiyev]

Would you like assistance consolidating some of these multi click tools into single tool call tools?

Let's see how #44 turns out.

@fanqiNO1 can get us posted as he develop a proper next-gen MCP calling experience. Hopefully #44 will:

• Reduce the get state calls by returning the state after actions, and
• Reduce the tokens used for each get state call by only including the updated fields

[CompleteDotTech]

That direction makes sense to me.

I have a very similar goal, and I'd be happy to put significant work into this project if we're aligned on where it should go. My current view is that the MCP surface is still a bit noisy, but I don't think the right end state is to hide game information from the agent. My intention is to give the agent the exact information a human has to play the game, just exposed in a structured way that is cheaper and easier to reason over.

So from my perspective the ideal direction is something like:

• fewer noisy / overly granular tool calls
• state returned with actions where it helps
• lighter-weight state updates or field-level diffs where possible
• but without losing the fidelity needed for good play and decision quality

If that's broadly in line with what you all want, let me know where you'd like help most and I'll put some real effort in. I'm happy to focus on whatever is highest leverage, whether that's MCP surface reduction, smarter state/update flows, or shaping the state contract so agents get the right information with less token noise.