Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Council Blackboard Create a proposal to increase the marketing budget and route it for approval."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Council Blackboard is an AI-first shared-state and multi-agent deliberation system with a Human-In-The-Loop (HITL) boundary, blackboard state, colloquium/chamber coordination, audit visibility, and a companion React control interface.
This repository is configured to deploy as an MCP server in Prefect Horizon.
Horizon entrypoint:
main.py:mcpHorizon detects pyproject.toml, installs fastmcp==4.0.3, imports the mcp object from main.py, and exposes the managed MCP endpoint.
Connect this GitHub repository to Prefect Horizon.
Create a server/project from aetheris-consulting/AI-council-blackboard.
Set the entrypoint to main.py:mcp.
Deploy and validate the tools in Horizon Inspector or ChatMCP.
See HORIZON.md for the exact deployment and local-validation commands.
Related MCP server: agent-broker
The Horizon-native server currently exposes blackboard state operations, HITL proposal routing and approval, operating-mode control, colloquium session inspection/interjection, colloquium-to-blackboard commits, rule-of-motion updates, resources, prompts, and a small in-process audit log.
The MVP is intentionally explicit about one limitation: Horizon state is currently in-memory and non-durable. It can reset on process replacement or redeploy. A durable shared-state backend should be added before multi-instance production use.
The original application remains in this repository as a React/Vite + Express TypeScript interface.
npm install
npm run devThe web application includes:
Council Blackboard UI
AI-first colloquium/chamber UI
HITL stage and proposal review
agent panels and rule-of-motion controls
connector and room UI
an earlier hand-written MCP JSON-RPC router in server/mcpServer.ts
The Horizon-facing main.py:mcp server is the recommended MCP deployment surface. The TypeScript router is retained for the companion application and future consolidation.
main.py # FastMCP server for Prefect Horizon
pyproject.toml # Horizon/Python dependency declaration
HORIZON.md # Horizon deployment notes
server/mcpServer.ts # Existing TypeScript MCP JSON-RPC implementation
server.ts # Express/Vite companion app server
src/ # React C2 / chamber UIThe Horizon-native MCP server requires no provider API key for its current core state-management tools.
The companion TypeScript application can use:
GEMINI_API_KEY=...Do not commit real credentials. .env* files remain ignored except for .env.example.
The source was reviewed before import for obvious embedded credentials. No live provider/API credential was found in the archive. The companion app does contain demo/session tokens and simulated/fallback behavior; those should not be treated as production authentication or durable governance controls.
Return recent mutating MCP calls from the in-process audit log.
Inputs: limit. Returns: result
Commit a new item directly to the shared blackboard.
Inputs: title (required), content (required), tags, type, +1 more. Returns: object
Approve or reject a pending proposal at the HITL supervisory boundary.
Inputs: action (required), proposal_id (required), feedback. Returns: object
Retrieve the shared Council Blackboard state.
Inputs: mode. Returns: object
Submit a candidate mutation through the Human-In-The-Loop boundary.
Inputs: title (required), content (required), agent_name (required), role, +2 more. Returns: object
Switch the operating mode of the Council Blackboard.
Inputs: mode (required). Returns: object
Commit a distilled colloquium outcome into the shared blackboard.
Inputs: title (required), content (required), tags, type. Returns: object
Inspect the current AI-first colloquium session.
Inputs: limit_messages. Returns: object
Insert a high-priority supervisor directive into the colloquium stream.
Inputs: directive (required), sender_name. Returns: object
Update colloquium turn structure, consensus threshold, or active topic.
Inputs: topic, rule_type, consensus_threshold. Returns: object
Return deployment identity, state model, and current chamber counts.
Inputs: no named parameters. Returns: object
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 11 tools
Some overlap exists: blackboard_add_item and colloquium_commit_to_blackboard both write to the blackboard, and get_state/get_session could be confused. However, descriptions clarify direct vs. distilled and state vs. session, so most tools have distinct purposes.
Most tools follow a <domain>_<action> pattern (e.g., blackboard_get_state, colloquium_interject), but 'system_status' lacks a verb, breaking the otherwise consistent naming scheme.
With 11 tools, the server is well-scoped, covering blackboard, colloquium, and audit operations without feeling bloated or sparse. This is within the ideal range for a specialized server.
The tool surface covers the core lifecycle of the domain: reading state, adding/proposing/approving items, managing sessions, and auditing. Minor gaps like explicit item deletion or update are not critical, so it feels reasonably complete.
Shared blackboard and MCP control plane for multi-agent teams.
Control plane for autonomous software labor. Agents claim objectives over MCP with audit trail.
Agent-native collaboration network: orchestrate a team of long-running agents from any MCP client.
Team agent collaboration: shared team facts, human-confirmed writes, tasks and threads over MCP.
