Overview

13 tool servers. 150+ tools. The stuff that makes AI agents actually useful.


How it works

When an agent needs to do something—search the web, create a diagram, query AWS—it calls a tool. The MCP proxy figures out which server handles that tool, checks permissions, and routes the request.


Servers

Server
What it does
Auth

awp-admin-mcp

System administration, audit logs, metrics

Admin only

awp-agenticode-mcp

Sandboxed code execution

User session

awp-aws-mcp

AWS operations via CLI

Azure AD OBO

awp-azure-mcp

Azure ARM API operations

Azure AD OBO

awp-azure-cost-mcp

Azure cost management

Azure AD OBO

awp-gcp-mcp

GCP API operations

Service account

awp-diagram-mcp

Diagram generation

None

awp-drawio-mcp

DrawIO diagram creation

None

awp-flowise-mcp

Flowise workflow management

User JWT

awp-servicenow-mcp

ServiceNow operations

Azure AD OBO

awp-web-mcp

Web search and fetch

None

awc-formatting-mcp

Content formatting

None


Tool naming

Tools follow the pattern {server}___{tool_name}:


Using tools

Via SDK

Direct execution

Via API


Auth

On-Behalf-Of (OBO)

Cloud tools use Azure AD's OBO flow. The user authenticates once, and the platform exchanges their token for cloud credentials. They get their own permissions, not a shared service account.

Admin RBAC

Admin tools check group membership. If you're not in the admin group, you can't use them.

Session isolation

Code execution tools create isolated workspaces per user. Your code runs in a sandbox.


Permissions

Permission
What it unlocks

can_use_web_search

awp_web.*

can_use_code_execution

awp_agenticode.*

can_use_image_generation

awp_diagram.*

flowise_enabled

awp_flowise.*

is_admin

awp_admin.*


Tool docs

Last updated