1. Conduid
  2. Communication
  3. core
MCP server · Communication

core

A terminal coding agent built on the Claude Agent SDK, with a web UI, agent browser automation, and MCP-based service connectors.

Unclaimed comms
44Fair

Scored 7 days ago · breakdown

About core

core is an MCP server published by git+Agentic-Engineering-Agency in the Communication category: a terminal coding agent built on the Claude Agent SDK, with a web UI, agent browser automation, and MCP-based service connectors.

The repository has 62 stars and 0 forks.

Install

Claude Code
claude mcp add agentforge -- npx -y agentforge-mcp
npx
npx -y agentforge-mcp

This server has no ConduID identity, so agent calls to it are not receipted. Pin the version you install and review the source before granting it credentials.

Ask AI

✦

Ask AI about core

Powered by Claude · Grounded in docs

I know everything about core. Ask me about installation, configuration, usage, or troubleshooting.

Security checks

  • ✓README presentThe repository ships documentation.
  • !Scoped permissionsDoesn't declare a permission scope. Assume it can do anything its process can.

4 checks have not been run yet.

README

AgentForge

AgentForge is a TypeScript monorepo for building and operating self-hosted AI agents with a persistent Mastra runtime, a Convex-backed data layer, and CLI-driven local workflows.

The repository currently contains the framework packages, the local Convex app used during development, the default project template scaffolded by the CLI, and example projects.

Monorepo Layout

agentforge/
├── packages/
│   ├── cli/                   # agentforge CLI and project scaffolding
│   ├── runtime/               # Persistent Mastra daemon and channel adapters
│   ├── core/                  # Shared agent, sandbox, MCP, channel, A2A, voice primitives
│   └── web/                   # Standalone dashboard package
├── convex/                    # Local Convex app copy used in repo development
├── templates/default/         # Synced default project template assets
├── examples/finforge/         # Example project
├── docs/                      # Project documentation
└── tests/e2e/                 # End-to-end test suite

The scaffolded project template source of truth lives under packages/cli/templates/default/, and Convex template changes must stay synced with the generated copies.

What Exists Today

  • @agentforge-ai/cli: scaffolds projects and provides the operational commands used during local development.
  • @agentforge-ai/runtime: exports the daemon, agent factory, Convex-backed memory helpers, and HTTP/Discord/Telegram channel adapters.
  • @agentforge-ai/core: exports shared agent primitives, sandboxing, MCP, channels, skills, A2A, research, workflows, voice, and browser tooling.
  • @agentforge-ai/web: a package-specific dashboard app.
  • convex/: the repo’s local Convex backend copy, which mirrors the default template’s Convex layer.

CLI Surface

The CLI currently registers these command families from packages/cli/src/index.ts:

  • Project lifecycle: create, run, deploy, upgrade
  • Runtime: start
  • Cloud/auth/config: models, workspace, tokens, config, vault, keys, status
  • Agent and chat: agents, chat, sessions, threads
  • Skills and orchestration: skills, skill, cron, workflows, research
  • Data and files: files, projects
  • Integrations: mcp, channel:telegram, channel:whatsapp, channel:slack, channel:discord
  • Execution: sandbox
  • Dashboard and logs: dashboard, logs are registered from the status command module

Important behavior:

  • agentforge run starts the local Convex development environment.
  • agentforge start boots the persistent daemon and HTTP channel.
  • agentforge chat talks to the daemon over the local HTTP/SSE endpoint.
  • agentforge dashboard starts a dashboard if it can find one in the project or repo layout.

Development

Prerequisites

  • Node.js >=18
  • pnpm >=8

Install Repo Dependencies

pnpm install

Standard Contributor Commands

Run these from the repo root while working on framework code:

pnpm build
pnpm test
pnpm typecheck
pnpm lint

When editing template-backed Convex files, sync the copies afterward:

pnpm sync-templates

Working On The Template Dashboard

packages/cli/templates/default/dashboard is not a workspace package. If you need to run or edit that dashboard directly, install its dependencies inside that directory:

cd packages/cli/templates/default/dashboard
pnpm install
pnpm dev

Start A Local AgentForge Project For Contribution

Use this flow when you need to validate the framework against a real project scaffold.

  1. Build the workspace packages from the repo root.
  2. Create a fresh project or use an existing scaffold.
  3. Configure the project .env.local.
  4. Start Convex.
  5. Start the AgentForge daemon.
  6. Chat with an agent and optionally start the dashboard.

Example flow:

# From the repo root
pnpm build

# Create a local test project
node packages/cli/dist/index.js create /tmp/agentforge-dev
cd /tmp/agentforge-dev

# Configure environment
cp .env.example .env.local

# Terminal 1: start Convex first
npx convex dev

# Terminal 2: start the daemon
agentforge start

# Terminal 3: talk to the daemon
agentforge chat

# Optional: start the dashboard if the project has a dashboard directory
agentforge dashboard

Environment And Startup Requirements

These requirements are enforced by the current command implementations:

  • CONVEX_URL must exist before agentforge start. The command reads it from the environment or from .env.local, and it expects npx convex dev to have been run first.
  • At least one agent must exist in Convex before agentforge start can boot usable runtime agents.
  • agentforge chat requires the daemon to already be running on the local HTTP port.
  • agentforge dashboard only starts if it can find a dashboard directory in the project, repo layout, or installed package path.

Documentation Map

Package Docs

Testing

  • Unit and package tests: pnpm test
  • Type checking: pnpm typecheck
  • End-to-end suite: see tests/e2e/README.md

License

Apache 2.0

README copy captured when this server was indexed; it may lag the repository. The original is authoritative.

Questions

About core

How do I install core?

Run claude mcp add agentforge -- npx -y agentforge-mcp, then add the server to your MCP client's configuration.

Is core safe to use with an AI agent?

Its trust score is 44 out of 100 (fair). It passes 1 of the 2 static security checks run so far; the failures are listed above. It has no ConduID identity yet, so agent calls to it are not receipted.

Is core still maintained?

Conduid hasn't recorded a commit date for this repository yet. Check the repository directly for recent activity.