Skip to content

Tools reference

Every tool an agent sees over MCP, what it does and its arguments.

Daemon administration. Actions: health, vectorize_session (session_id), vectorize_stats, create_checkpoint (session_id).

Arguments: action (required), session_id.

Everything for one task in one call, within token_budget: the user and project cards first, then the graph entities named in the query with their one-hop neighbourhood and what depends on them (at most half of what the cards leave), then the records that best match the query, ranked by the same search as semantic_search (only records that still hold, each shown as its passage for the query). With sources=messages or all, also adds captured conversation turns matching the query as excerpts. Returns ranked items, entity neighbourhood, dependency impact, and an LLM-ready formatted block. Scope: session or workspace.

Arguments: query (required), select, select_map, session_id, sources, token_budget, workspace_id.

Get session summary. Use ‘include’ to specify sections: decisions, insights, entities, or all (default).

Arguments: compact, concepts_limit, decisions_limit, entities_limit, include, min_confidence, questions_limit, session_id (required).

The documents modules keep in a workspace, versioned. Actions: put (kind, data, text, expected_version; document_id for a change), get, history, list (kind, fields, limit, cursor), delete. Only the module that declares a kind writes it; anyone who reads the workspace reads it.

Arguments: action (required), author, cursor, data, document_id, expected_version, fields, kind, limit, text, version, workspace_id (required).

A session’s message stream, its conversation turns. Actions: append (messages: each with turn_id, role, text, and for an agent’s turn its parent and blocks), page (from, limit; oldest first, next_from for the next page). Agents keep their turns whole here to resume them.

Arguments: action (required), from, limit, messages, session_id (required).

Manage entities and context entries in a session. Actions: delete (remove entity + cascade typed edges; requires entity_name), delete_update (remove a single context entry by entry_id).

Arguments: action (required), entity_name, entry_id, into, session_id (required).

Manage workspaces and their cards. Actions: create, list, get, delete, add_session, remove_session; get_card / update_card for the project card (workspace_id) or the user card (card: “user”): short markdown notes put in front of every session

Arguments: action (required), author, card, content, description, expected_version, history, include_archived, key, kind, name, role, session_id, workspace_id.

Query session data. Types include: find_related_entities, get_entity_context, search_updates, entity_importance, entity_network, find_related_content, key_decisions, key_insights, session_statistics, structured_summary, decisions, open_questions, assemble_context, record (one record whole by entry_id, e.g. after a search snippet).

Arguments: parameters (required), query_type (required), session_id (required).

Universal semantic search. Scope: session (requires scope_id), workspace (requires scope_id), or global (default).

Arguments: author, date_from, date_to, detail, interaction_type, limit, mode, prefer_author, query (required), recency_bias, scope, scope_id, select, select_map, sources, trace.

Manage sessions. Actions: create, list, load, search, update_metadata, delete; resolve_project (key, name, optional workspace) finds or creates the workspace and session of a project.

Arguments: action (required), description, key, name, query, seq, session_id, workspace.

Add context updates to conversation. Supports single update or bulk mode with updates array.

Arguments: author, code_reference, confidence, content, derived_from, dry_run, entities, interaction_type, relations, session_id (required), supersedes, updates, valid_from.