MCP servers for Omilia Cloud Platform
概览
A Model Context Protocol server that lets MCP-aware LLM clients build, deploy, and operate conversational AI agents on the through natural language. Create an orchestrator app, spin up an autonomous agent, attach a knowledge base from a URL, wire a webservice tool, deploy, and smoke-test the result — all from a chat prompt in your editor or desktop app. The server runs locally and talks to OCP over authenticated HTTPS. No data leaves your machine except the requests your MCP client makes on your behalf. Claude Desktop, Claude Code, Cursor, VS Code (with Copilot), and Codex. The wizard selects your client, collects credentials with masked input, runs prereq checks (Node 20+, uv), installs the config, and prints a PASS/FAIL smoke-test summary before exiting. See the Interactive wizard section in the installation guide for the full prompt sequence. See the full installation guide at docs/installation.md for the .
README
以下为原文 README 节选，完整内容请查看原文。
Omilia MCP Server
Documentation: learn.ocp.ai
A Model Context Protocol server that lets MCP-aware LLM clients build, deploy, and operate conversational AI agents on the Omilia Cloud Platform (OCP) through natural language. Create an orchestrator app, spin up an autonomous agent, attach a knowledge base from a URL, wire a webservice tool, deploy, and smoke-test the result — all from a chat prompt in your editor or desktop app.
The server runs locally and talks to OCP over authenticated HTTPS. No data leaves your machine except the requests your MCP client makes on your behalf.
Supported MCP clients
Claude Desktop, Claude Code, Cursor, VS Code (with Copilot), and Codex.
Install
Recommended — interactive wizard (answers a few prompts, then confirms before writing anything):
npx github:omilia/mcp init
The wizard selects your client, collects credentials with masked input,
runs prereq checks (Node 20+, uv), installs the config, and prints a
PASS/FAIL smoke-test summary before exiting. See the
Interactive wizard section
in the installation guide for the full prompt sequence.
Non-interactive / CI — supply all flags to skip prompts:
npx github:omilia/mcp init --client
See the full installation guide at docs/installation.md
for the .mcpb one-click bundle (Claude Desktop) and manual configuration
alternatives.
Authentication
Recommended: Keycloak password grant — set OCP_USERNAME, OCP_PASSWORD, and OCP_KEYCLOAK_REALM (default master).
Faster path: a Personal Access Token via OCP_ACCESS_TOKEN.
Both run in-memory only — credentials never persist. See docs/installation.md for the realm-by-environment table.
Tools (31)
The MCP surface, grouped by area. Each row is one tool your MCP-aware
LLM can invoke through tools/call.
Discovery & guides
| Tool | Description |
|---|---|
read_guide |
Return a markdown guide for a canonical OCP MCP workflow |
list_groups |
List OCP groups the current user has access to |
list_agents |
List all agents in scope |
list_knowledge_bases |
List knowledge bases (FAQ vector stores) |
list_pathfinder_projects |
List Pathfinder projects (search filterable) |
search_orchestrator_apps |
Search Orchestrator apps |
search_miniapps |
Search miniapps |
search_numbers |
Search phone numbers attached to apps |
search_variable_collections |
Search variable collections |
search_dialog_logs |
Search dialog logs by group / app / date |
Orchestrator apps
| Tool | Description |
|---|---|
create_orchestrator_app |
Create a new Orchestrator app |
get_orchestrator_app |
Fetch an app’s canvas |
add_concierge_to_orc_app |
Wire a Concierge into an app’s canvas |
deploy_orc_app |
Deploy an Orchestrator app |
安装
npx github:omilia/mcp init