home / mcp / ordinal-mcp mcp server
Provides an ordinal, multi-level message bus for structured AI-human coordination and decision making.
Configuration
View docs{
"mcpServers": {
"bobbyhiddn-ordinal-mcp": {
"command": "uv",
"args": [
"run",
"--directory",
"/path/to/Ordinal-MCP",
"ordinal-mcp"
]
}
}
}Ordinal-MCP provides a structured ordinal bus that coordinates communication between AI agents and human operators. It enables upward oracle calls and downward task delegation with typed, archived messages to keep interactions clear and auditable. This server is built to support controlled human-in-the-loop decisions within agent systems.
You will run the MCP server locally or in your infrastructure and connect it to an MCP client. Use the provided command-line tools to send oracle questions, monitor pending calls, and review completed exchanges. When a subagent needs guidance, Rhode answers via its own LLM. When Rhode needs human input, the system escalates to the human oracle through Telegram, and the answer flows back to the requester.
Key usage patterns include: initiating oracle calls from different ordinal levels, checking bus status to see pending and completed exchanges, and inspecting history to audit past decisions. Use the standard JSON-based bus directory to inspect requests, responses, and archived conversations. If a subagentβs query requires human judgment, Rhode will indicate that in its answer and the subagent can re-ask with a higher from_level to escalate.
Prerequisites: Python 3.13 or newer and the uv runtime for the command-line interface.
1) Clone the project repository.
2) Change into the repository directory.
3) Install and start the MCP runtime. You can synchronize dependencies with uv and prepare the environment as shown.
4) Start the standalone server for testing.
The server runs as an MCP server configured to be started via a standard stdio command. Use the following approach to connect to an MCP client in your project and define the ordinal bus interaction: add a stdio MCP server entry pointing to the runtime and the ordinal-mcp module, then start the orchestration.
The Ordinal-MCP provides a set of tools to interact with the bus and manage oracle calls. You can send questions, check current pending calls, respond to oracle calls, and view exchange histories. The available tools are listed below with their purposes and usage notes.
- Ensure the bus directory is writable by the running process so requests, responses, and history can be written and archived correctly.
- When escalating to the human oracle, make sure the Telegram integration is configured and the human responder is ready to reply with the required format.
Send a question to the appropriate oracle and wait for a response, handling from_level, to_level, and context parameters.
Check the current state of the bus, including pending requests, responses, and history counts.
List all oracle calls that are currently waiting for a response.
Provide an answer to a pending oracle call, updating the bus with the responder and timestamp.
View recent completed exchanges with optional limits to control how many records are shown.