# Jovancoding/Network-AI [Health: Active]

**Category:** 🤖 Coding Agents  
**Repository:** https://github.com/Jovancoding/Network-AI  
**GitHub Stars:** 79  
**npm Downloads (last month):** 17  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/jovancoding-network-ai

## Description
Multi-agent orchestration MCP server with race-condition-safe shared blackboard. 20+ MCP tools: blackboard read/write, agent spawn/stop, FSM transitions, budget tracking, token management, and audit log query. npx network-ai-server --port 3001.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "network-ai": {
    "command": "npx",
    "args": ["-y","network-ai-server"]
  }
}
```

## Documentation

## What Jovancoding/Network-AI MCP server does

Jovancoding/Network-AI MCP server provides a coordination layer for multi-agent applications. Its central shared blackboard lets agents exchange project state while using an atomic propose, validate, and commit flow. That flow is designed to prevent concurrent writes from silently overwriting one another or leaving agents with inconsistent views of shared data.

The server also applies operational controls around agent activity. These include finite-state-machine governance, per-agent token ceilings, permission gating, signed audit trails using HMAC or Ed25519, and budget tracking. Agents can be started or stopped, moved through defined states, and monitored through audit queries.

A broader Network-AI installation can coordinate agents from supported frameworks through adapters, including LangChain, AutoGen, CrewAI, OpenAI Assistants and Responses, LlamaIndex, Semantic Kernel, Haystack, DSPy, Agno, LangGraph, OpenAI Agents SDK, Vertex AI, Gemini, Pydantic AI, and others listed by the project. The MCP interface is most relevant when an MCP client or agent needs access to the coordination control plane.

## How it works

Agents interact with shared state through MCP operations rather than writing directly to a common value without coordination. A proposed change is checked before it is committed, providing a synchronization point for parallel work. The blackboard can also be searched, and the `context_pack` capability can assemble a bounded, relevance-ranked selection of state instead of requiring an agent to load the entire board.

Governance features cover more than state changes. Budget and token controls constrain usage, while the audit system records activity for later inspection. The project also includes lifecycle governance for model interactions in its wider TypeScript/Node.js package, including fallback handling, model budget accounting, refusal telemetry, effort policy, and thinking-block management. These package-level capabilities should not be assumed to be separate MCP tools unless exposed by the running server configuration.

## Setup and configuration

Jovancoding/Network-AI MCP server can be launched with:

```bash
npx network-ai-server --port 3001
```

The README identifies Node.js 18 or newer as a requirement. The command selects port 3001 for the server process. Network-AI is also available as a TypeScript/Node.js library and command-line interface, but those are alternative usage modes rather than prerequisites for starting the MCP server.

The repository identifies the project as MIT-licensed. Its documentation also describes HTTP and gRPC transport modules in the wider project, but the supplied server command alone does not specify which transport is selected in every deployment.

## Tools and capabilities

- Reading and writing blackboard state
- Searching shared context and creating context packs
- Starting and stopping agents
- Performing finite-state-machine transitions
- Tracking budgets and tokens
- Querying the audit log

Jovancoding/Network-AI MCP server is therefore suited to workflows where several agents must share decisions, coordinate progress, and operate within explicit resource or permission limits. It is less directly suited to a single-agent setup that only needs an ordinary API wrapper or an isolated task tool.

## Limitations and notes

The shared-blackboard protection addresses coordination at the control-plane layer; it does not by itself make every external system transaction atomic. The project describes inter-agent communication as only partially covered within its stated OWASP Agentic AI mapping because the local trust boundary remains relevant.

The README names 32 adapters and several standalone usage modes, but adapter availability does not mean every framework is automatically connected to an MCP client. Framework-specific setup may require using the library or the relevant adapter documentation.

_Full upstream README: https://allmcps.com/mcp/jovancoding-network-ai/readme_