grimoire-campaign-context

Working with Grimoire

Grimoire is a campaign manager for tabletop RPG gamemasters, and its MCP server gives you live read and write access to one campaign's database. That database is not a pile of notes. It is typed: 14 entity categories, typed relationships between them, three knowledge graph projections, open narrative threads, and a block-based wiki whose pages link to the records. The endpoint is https://api.ttrpg.bot/mcp, transport is streamable HTTP, auth is OAuth 2.1, and access is scoped to the single campaign the user approved.

Your job is to answer from that campaign's canon, not from generic tabletop RPG knowledge. The homebrew is the authority. If the campaign says elves are extinct and magic costs blood, that is true, and your training data is wrong.

Grimoire exposes 52 tools in 8 groups. The tools carry their own layer numbering in their descriptions. Follow it. Every tool description is current; when this skill and a tool description disagree, the tool description wins.

Center yourself first

Do not answer a campaign question, and do not write anything, until you have grounded yourself in this campaign. The whole point of the layered design is that context gets loaded deliberately instead of dumped all at once.

Every session, before the first substantive answer:

1. current_campaign (free, instant). Returns the campaign, user, and role bound to this MCP session. Call it first, every time. Everything after this depends on it, and the role decides how much of the next two steps you get back. A player token also lists the player's own characters under characters. See "If you are connected as a player" below.
2. get_constitution (Layer 1). Both roles. The world's non-negotiable truths: campaign summary, Campaign Bible summary, and the world foundations graph. This is where the setting's defining events, cosmology, and the GM's content boundaries live. It colors every downstream answer. A player-role connection gets its table's own constitutional layer, never the GM's: the response carries a shared object (bible, foundations, and a graphs map), and the Bible summary and foundation nodes are present only where the matching flag is true. Read shared first, because a false flag means the GM did not share that part, not that the call failed.
3. The campaignContext block. Genre, ruleset, setting technology and magic levels, which entity categories this campaign uses, and what this campaign calls them. It rides inside the get_constitution response you just read, and inside get_narrative_state and get_entity_catalog as well, so read it there instead of fetching it again. get_campaign_context is the standalone form of the same block, for a session that needs nothing but the vocabulary. Read it either way: it is how you stop saying "NPC" when this table says "contact," and how you stop applying D&D 5e assumptions to a Blades-adjacent horror game.