MCP tool reference

Every tool, grouped by what it does.

All 34 tools, grouped by purpose. Call tools/list on the MCP endpoint for full JSON schemas — that is the authoritative contract and what you should build tool definitions from.

Two rules that save time. Ids are never guessable — get neuron type ids from aiqbee_get_neuron_types before writing. And every tool returns errors as structured JSON with actionable text; pass that back to the model rather than collapsing it to "call failed".

Search & read

The tools your agent will call most. `aiqbee_search` is semantic — pass a natural-language question, not keywords.

Tool	What it does
`aiqbee_search`	Semantic search across neurons in the brain.
`aiqbee_fetch`	Retrieve one neuron in full by id.
`aiqbee_list_neurons`	Page through neurons, optionally filtered by type.
`aiqbee_get_neuron_count`	Count neurons — cheap way to size a brain.
`aiqbee_get_brain_info`	Brain name, description and settings.
`aiqbee_get_relationships`	Synapses (links) for a given neuron.
`aiqbee_get_neuron_types`	Types available in this brain, with ids.
`aiqbee_list_neuron_types`	Same set, listing form.
`aiqbee_list_metadata_fields`	Metadata fields this brain declares — the keys usable in <code>metadata_filter</code> on search and list.
`aiqbee_edit_is_allowed`	Check writability before offering an edit.

Write

All writes respect the brain's allow_mcp_editing flag and the caller's access level. By default, creating a neuron whose title and content exactly match an existing one is rejected, and the existing neuron is returned so the agent can update or link instead of fragmenting the brain; pass allow_duplicate=true only when a second copy is genuinely intended.

Tool	What it does
`aiqbee_create_neuron` writes	Create a neuron.
`aiqbee_update_neuron` writes	Update a neuron's title, content or date.
`aiqbee_delete_neuron` writes	Delete a neuron.
`aiqbee_create_relationship` writes	Link two neurons with a synapse.
`aiqbee_update_relationship` writes	Edit a synapse's description.
`aiqbee_delete_relationship` writes	Remove a synapse.
`aiqbee_update_brain` writes	Change brain settings.
`aiqbee_add_neuron_type` writes	Add a neuron type.
`aiqbee_edit_neuron_type` writes	Rename or re-describe a type.
`aiqbee_delete_neuron_type` writes	Delete a type, reassigning its neurons.

Access control

Directory and grant management. The per-type grant tools require Advanced Security, which is a paid feature — see Licensing & tiers.

Tool	What it does
`aiqbee_list_users`	Users with access to this brain.
`aiqbee_list_all_groups`	All user groups on this Hive.
`aiqbee_list_brain_groups`	Groups with access to this brain.
`aiqbee_list_all_service_accounts`	Service accounts on this Hive.
`aiqbee_list_type_access`	Per-type grants for a neuron type.
`aiqbee_grant_access` writes	Grant a user access to the brain.
`aiqbee_revoke_access` writes	Revoke a user's brain access.
`aiqbee_grant_brain_group_access` writes	Grant a group brain access.
`aiqbee_revoke_brain_group_access` writes	Revoke a group's brain access.
`aiqbee_grant_type_access` writes	Grant access to one neuron type.
`aiqbee_revoke_type_access` writes	Revoke access to one neuron type.
`aiqbee_set_neuron_type_security` writes	Opt a neuron into type-level security.

Audit history

Audit logging is a paid feature and must be switched on per brain. These tools refuse when it is off.

Tool	What it does
`aiqbee_get_audit_history`	Change history for one neuron.
`aiqbee_get_brain_audit_history`	Change history across the brain.

Choosing which to expose

You do not have to hand your model all 34. A larger toolset makes selection harder, so start narrow and widen when you see a real need:

Agent shape	Reasonable toolset
Answers questions from team knowledge	`aiqbee_search`, `aiqbee_fetch`
Also records what gets decided	the above plus `aiqbee_get_neuron_types`, `aiqbee_create_neuron`, `aiqbee_update_neuron`
Curates and connects knowledge	the above plus `aiqbee_create_relationship`, `aiqbee_get_relationships`
Administers the Hive	the access-control group — usually better done by a human in the UI

Leaving the delete tools out entirely is a reasonable default. Deletion is rarely what a conversational agent should be doing unprompted, and an accidental one is expensive to notice.