SkillFoxx

Network-AI MCP

A TypeScript multi-agent orchestrator with a locking shared blackboard, token budgets and an audit trail.

  • 89 ★ on the repository
  • Last commit
  • MIT
MCP server

Extended access

Why this level

  • Writes shared agent state and an audit log to local project files
  • The model gateway and adapters send requests to the chosen providers
All reasons and checks

jovancoding/network-aiAuthor:Jovancoding

Install

In your terminal, with SkillFoxx CLI

npx skillfoxx add mcp/network-ai

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Extended accessThe tool runs code or makes network calls and only reads external data.

  • Writes shared agent state and an audit log to local project files
  • The model gateway and adapters send requests to the chosen providers

This is third-party code. Review the repository files before installing.

What it does

Network-AI adds coordination and control to a stack of several AI agents. A locking shared blackboard sends every write through a propose, validate, commit cycle so parallel agents do not overwrite each other's keys. Alongside it run per-agent token ceilings, a finite-state permission machine, a signed audit trail (HMAC and Ed25519) and permission gating for actions. The project works as a TypeScript library, an MCP server, a CLI, a Claude Code plugin, a Gemini CLI extension and an OpenClaw skill. The MCP server has `context_pack` and `blackboard_search` tools: an agent pulls relevance-ranked context instead of the whole board. The README lists 32 adapters, including LangChain, AutoGen, CrewAI, Claude Agent SDK and OpenAI Agents SDK.

Who it is for. Teams that run several agents in parallel and hit overwritten shared state and uncontrolled spending.

Good fit when

  • Several agents write to one shared state and results overwrite each other without an error.
  • Each agent needs a token ceiling and a dollar budget so the chain does not exceed a limit.
  • Agent actions need a signed audit trail for post-incident review.
  • Agents from different frameworks need one state board, and glue code per pair is unwelcome.

Not a fit when

  • A single agent works with no parallel writes: the orchestrator adds needless complexity.
  • A ready-made cloud service is wanted: Network-AI runs on your side, on Node.js.
  • Only a skill for one task is needed: the project is closer to infrastructure than to a single ability.

Example request

Set up a shared blackboard for three agents, give each a token ceiling and show the audit log after the run.

MCP

Transport
http
Authentication
not required

Limitations

The MCP server needs Node.js: start it with `npx -y -p network-ai@6.0.1 network-ai-server --port 3001`. The OpenClaw skill bundle contains only local Python scripts, and the TypeScript engine is not part of that package. The project is large, with 32 adapters, record and replay modules, job queues and a model gateway, so learning it takes time. `GovernedModelGateway` and budget policies must be configured for your own models. The version in the launch command comes from the README and may go stale.

How to disable. Stop the server and remove the extension with the client command, for example `/plugin uninstall network-ai`. Delete `swarm-blackboard.md` and the `data/` directory in the project by hand.

FAQ

What problem does the shared blackboard solve?

It stops parallel agents from silently overwriting one key. A write goes through a propose, validate, commit cycle under a lock.

How do I connect it to Claude Code?

With `/plugin install network-ai@network-ai`. For Gemini CLI there is `gemini extensions install https://github.com/Jovancoding/Network-AI`.

How do I use it from the command line?

With commands such as `network-ai bb get status` for the board and `network-ai audit tail` for the audit log. Other commands are in the project documentation.

Which frameworks does it work with?

The README names LangChain, AutoGen, CrewAI, LlamaIndex, Semantic Kernel, Haystack, DSPy, Claude Agent SDK and OpenAI Agents SDK. There are 32 adapters in total.

What is in the OpenClaw skill?

Local Python scripts for the blackboard, permission checks, token budgeting and project context. They make no network calls, and the skill has no TypeScript engine.

Discussion

No questions or reviews yet.

MCP servers

A plugin that gives an agent the IntelliNode skill and MCP server: ask any model, search, review code, tests, SQL, images and speech

PluginBroad access275

Self-hosted system for a team of AI agents: org chart, tasks, budgets and oversight of Claude Code, Codex and OpenClaw

CLIBroad access113.3K

A hub and the anet CLI join Claude Code, Codex, Grok and OpenCode from different machines into a network driven from one window.

CLIBroad access120

A bridge between local CLI agents: Codex, Cursor or Claude Code delegate work to Antigravity, Grok, Kimi and seven more.

MCP serverBroad access132

A local dashboard for running many agents across machines: memory, code graph and agent-to-agent messaging

CLIBroad access852

A CLI that keeps instructions, commands, skills and MCP in .agents and links them into the configs of different agents

CLIExtended access76

README in short

Covers orchestrator capabilities, usage as a library, MCP server, CLI and plugin, the adapter list and model interaction governance.

README badge

Report a problem