Skip to main content

​
MCP Tools Reference

The hosted MCP server exposes tools built from AUVY’s shared metadata plus Connect-managed integration tools. The exact live tool list comes from the server, but these are the main tool families you should expect.

​
Session Context

​
get_session

Returns connection context for the authenticated user:
  • workspace_id
  • workspace_slug
  • brain_id
  • default_brain_id
Use this first to verify workspace/brain scope.

​
Discovery Tools

​
grep

Keyword browse/search using query in type/pattern format (for example neurons/support). Semantic search using query in type/pattern format.

​
read / cat

Read resource details by path (or paths). cat is an alias of read.

​
grep_history / read_history

Inspect prior execution history for the current neuron and brain context.

​
list_receptors

List receptors (invokable entry points). Optional: target_type, limit, offset.

​
list_pathways

List pathways. Optional: status, visibility, search, limit, offset.

​
list_neurons

List neurons. Optional: visibility.

​
list_reflexes

List reflexes. Optional: type, visibility, search, limit, offset.

​
list_neuron_reflexes

List reflexes for a neuron. Required: neuron_id.

​
Web Tools

​
web_read

Fetch page content from one or more URLs. Run one or more web searches.

​
Creation And Update

​
write

Create new resources with a path + description workflow:
  • pathways/<name>
  • neurons/<name>
  • reflexes/<name>
  • receptors/<slug>

​
write_reflex

Alias focused on creating reflexes.

​
edit

Update existing resources by path with config.

​
create_resource / store

Lower-level creation helpers used for resource creation and artifact or memory storage.

​
Execution

​
invoke

Invoke a receptor or neuron by path. For public assistant usage, this is the main execution tool to know about.

​
connect_execute

Execute a Connect integration tool directly. Auvy uses direct execution: you choose which tools to run and call connect_execute with the integration and tool name (no session or meta-tool discovery). Required input:
  • integration_id (UUID)
  • tool_name
Optional input:
  • tool_config (object) — tool arguments
  • connected_account_id — which connected account to use when the user has multiple
Use connect_execute with an integration that has already been configured in the AUVY dashboard. This docs set does not cover the separate JWT-only integration-management surface.

​
Skills

​
learn_skill

Load a skill by skill_id (UUID). Returns name, description, and body for use at runtime.

​
create_skill

Create a skill from body_md (markdown). Returns skill id for use with learn_skill.

​
remove_neuron_reflex

Remove a neuron-reflex association. Required: neuron_id, reflex_id.

​
MCP App UI

The read tool can return an MCP App resource for receptors:
  • ui://receptor-viewer/mcp-app.html
The current MCP app is implemented in Vue (apps/mcp/resources/receptor-viewer).

​
Brain Context

Some tools are brain-scoped. Set X-Brain-Id in MCP headers when needed: 
{
  "headers": {
    "Authorization": "Bearer ak_live_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
    "X-Brain-Id": "00000000-0000-0000-0000-000000000000"
  }
}