MCP Servers
24 Model Context Protocol servers providing 140+ tools to the agent.
Core Tools core
| Server | Tools | Description |
|---|
| git | 9 | Status, diff, commit, branch, push, pull, stash, blame, log |
| workspace | 5 | Project info, file listing, git status, port detection |
| web | 4 | HTTP fetch, search, documentation reader, API calls |
| browser | 10 | Navigate, click, type, extract, screenshot (Playwright) |
| secrets | 7 | AES-256-GCM vault, set/get/rotate/audit keys |
| tasks | 5 | Create, list, update, search, delete project tasks |
| lsp | 6 | Language server diagnostics, formatting, completions |
| plugins | 5 | Install/manage extensions from git or npm |
Team team
| Server | Tools | Description |
|---|
| team | 5 | Status, knowledge sharing, blockers, team context |
| notifications | 9 | Slack, Discord, webhook alerts |
| messaging | 8 | Telegram, Discord, Slack, webhook messaging gateway |
AI & Learning ai
| Server | Tools | Description |
|---|
| memory | 9 | Capture, recall, deep recall, session gists, team context |
| model-router | 6 | Multi-provider routing, cost optimization, MoA |
| metrics | 7 | Usage stats, cost tracking, token analytics |
| compaction | 3 | Context compression, topic threads, auto-triggers |
| user-modeling | 6 | Expertise profiling, adaptive behavior |
| skill-evolution | 7 | Bayesian skill quality scoring, A/B testing |
| trainer | 5 | Training scenarios, scoring, trajectory export |
Operations ops
| Server | Tools | Description |
|---|
| hooks | 5 | Pre/post tool hooks, policy enforcement, audit logging |
| permissions | 5 | RBAC, tool-level access control |
| scheduler | 7 | Natural language cron scheduling via NATS |
| simulator | 5 | Test runner, benchmarks, isolated execution |
| harness | 5 | Multi-agent DAG execution with roles |
| mcp-gateway | 3 | MCP protocol gateway and routing |
How MCP works
Each MCP server is a separate process that communicates with the agent via JSON-RPC over stdio. Servers register tools that the AI can call. When the agent needs to perform an action (like reading a file or running a git command), it invokes the appropriate tool on the relevant MCP server.
All 24 servers start automatically when your workspace boots. You can add custom MCP servers by editing ~/.arcx/mcp.json.