AI MCP Servers

ⓘ Open AI → MCP Servers. See also the AI Assistant and Model Platform.

Permissions

Browse the catalog with mcp:use. Run tools from the UI or Assistant with ai_gateway:use plus each tool’s requiredPermission. Approve or reject HIGH-risk executions with ai_gateway:approve on AI → Activity → Tool runs.

What it is

Axio exposes Model Context Protocol–style servers (domains) and tools over HTTP in two layers:

1

Platform MCP — 22 org-scoped domain servers bound to real platform services. Each tool declares requiredPermission, parameter validation, audit logging, and a risk level (LOW, MEDIUM, HIGH). HIGH-risk tools pause for human approval before execution completes.

2

Classic MCP — JSON-RPC 2.0 Streamable HTTP, plus a human-readable manifest. Protocol version 2025-06-18, server name axio-mcp. The manifest merges classic tools with all Platform MCP tools (qualified names such as projects.list_projects).

♢ The MCP Servers page loads Platform MCP discovery and the Copilot tool-calling API — the same execution records as the Assistant. The UI lists all Platform MCP tools; the Assistant system prompt only recommends tools marked copilotExpose.

How to use

1 ⌕

Browse servers by domain; search and filter by risk. Inspect tool name, qualified name, parameters, required permission, and risk badge.

2 ▣

Run from the catalog form, use Ask Copilot to open the Assistant with a prefilled prompt, or name a tool in natural language in chat.

3 ♢

Review results in the run dialog. Approve or reject HIGH-risk runs on AI → Activity → Tool runs or via quick-action badges when pending approvals exist.

Catalog features

✓ Domain-grouped tool table with server metadata from discovery

✓ Risk filter (ALL / LOW / MEDIUM / HIGH) and text search

✓ Favorites, tool details drawer, and recent execution history on the page

✓ Project/workspace pickers for tools that accept hierarchy refs

✓ Quick actions: Open Assistant, Tool runs, Pending approvals

✓ Every tool enforces requiredPermission at execution time. HIGH riskLevel tools enter AWAITING_APPROVAL until a user with ai_gateway:approve approves or rejects. LOW and MEDIUM tools run immediately after RBAC and input validation (including secret scanning on string inputs).

Classic MCP tools

Built-in JSON-RPC tools:

▣ plan Terraform/OpenTofu plan for a workspace

▱ chat Legacy copilot chat message

✓ apply Apply workspace

⟳ rollback Previous Terraform state version

▱ destroy Destroy workspace resources

⌕ search Search projects, stacks, workspaces, policies, variables

⌁ drift Run a drift monitor check

✦ generate Generate Terraform or Pulumi from a prompt

♢ approve / cancel Approval request decisions

▥ analyze Architecture / cost / security analysis for a workspace