Skip to content
/ Jarvis-mcpm Public

Next-Gen MCP Gateway. Give your AI agent the ability to dynamically manage its own MCP tools.

2 stars 0 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

JRedeker/Jarvis-mcpm

BranchesTags

Repository files navigation

Jarvis

     ██╗ █████╗ ██████╗ ██╗   ██╗██╗███████╗
     ██║██╔══██╗██╔══██╗██║   ██║██║██╔════╝
     ██║███████║██████╔╝██║   ██║██║███████╗
██   ██║██╔══██║██╔══██╗╚██╗ ██╔╝██║╚════██║
╚█████╔╝██║  ██║██║  ██║ ╚████╔╝ ██║███████║
 ╚════╝ ╚═╝  ╚═╝╚═╝  ╚═╝  ╚═══╝  ╚═╝╚══════╝

The Agentic MCP Server for Dynamic Tool Management

Go License MCP Tests Go Report

Jarvis lets your AI agent manage its own tools. Install servers, switch profiles, configure clients—all through natural language. One MCP server to rule them all.

The Problem

You're using Claude, Cursor, or another AI client. You have MCP servers for memory, search, code tools. But:

  • Static configs — Every new tool requires manual JSON editing and client restart
  • Per-client duplication — Same servers configured separately in Claude, Cursor, VS Code
  • No agent autonomy — Your agent can't install tools it needs mid-conversation
  • Context switching — Different projects need different tool sets

The Solution

Jarvis is an agentic-first MCP server that gives your AI agent control over its own tooling:

You: "I need to analyze some PDFs"
Agent: [searches registry → installs pdf-parse → uses it immediately]
Agent: "Done. The contract has a 30-day payment term on page 3."

No config editing. No restart. The agent handles it.

Key Features

  • Dynamic Tool Installation — Install MCP servers mid-conversation from 200+ registry
  • Profile-Based Tool Sets — Composable micro-profiles for different workflows
  • Multi-Client Management — Configure Claude, Cursor, VS Code from one place
  • Self-Healing Infrastructure — Agent detects and repairs its own tooling
  • Context-Efficient — 52% smaller tool definitions (~1,400 tokens saved per connection)

Core Capabilities

1. Dynamic Tool Installation

Your agent discovers and installs tools from a registry of 200+ MCP servers:

jarvis_server({ action: "search", query: "pdf" })     // Find tools
jarvis_server({ action: "install", name: "pdf-parse" }) // Install
// Tool is immediately available—no restart needed

2. Profile-Based Tool Sets

Group tools into composable profiles. Switch entire toolsets based on project context:

jarvis_profile({ action: "suggest" })  // Auto-detect best profile for current directory
// Returns: "essentials,dev-core,research" for coding projects
Profile Tools Use Case
essentials Time, Fetch Always-on utilities
dev-core Context7 Coding intelligence
research Kagi, Firecrawl Web research (Docker)
memory Basic Memory Persistence
data Qdrant, Postgres Heavy DBs

3. Multi-Client Management

Configure Claude Desktop, Cursor, VS Code, OpenCode—all from one place:

jarvis_client({ action: "edit", client_name: "cursor", add_profile: "memory" })
// Cursor now has memory tools. No manual JSON editing.

4. Self-Healing Infrastructure

Agent detects and repairs its own infrastructure:

You: "My search isn't working"
Agent: [checks status → finds Qdrant down → restarts → confirms healthy]
Agent: "Fixed. Qdrant was down, restarted it. Search should work now."

Installation

git clone https://github.com/JRedeker/Jarvis-mcpm.git
./Jarvis-mcpm/scripts/setup-jarvis.sh

Copy the output JSON into your AI client config, or use --auto-config to do it automatically.

How It Works

flowchart TB
    subgraph Clients["AI Clients"]
        Claude[Claude Desktop]
        Cursor[Cursor]
        VSCode[VS Code]
        OpenCode[OpenCode]
    end

    subgraph Jarvis["Jarvis (Agentic MCP Server)"]
        Gateway[Tool Management API]
        Registry[(200+ Server Registry)]
    end

    subgraph Daemon["MCPM Daemon"]
        P1[Profile: Research]
        P2[Profile: Memory]
        P3[Profile: Dev-Core]
    end

    Clients -->|"Install/Switch/Configure"| Gateway
    Gateway --> Registry
    Gateway -->|Manage| Daemon
    Daemon --> P1 & P2 & P3
Loading

Jarvis sits between your AI clients and the MCP ecosystem. It's the only MCP server your agent needs to manage all other MCP servers.

Tool Reference

9 consolidated tools using action-based routing — 52% smaller than traditional definitions (~1,400 tokens saved per connection):

Tool Actions Purpose
jarvis_check_status — System health check
jarvis_server list, info, install, uninstall, search, edit, create, usage MCP server management
jarvis_profile list, create, edit, delete, suggest, restart Profile management
jarvis_client list, edit, import, config AI client configuration
jarvis_config get, set, list, migrate MCPM settings
jarvis_project analyze, diff, devops Project analysis & DevOps
jarvis_system bootstrap, restart, restart_infra System operations
jarvis_share start, stop, list Server sharing
jarvis_diagnose profile_health, test_endpoint, logs, full, config_sync MCP debugging
Example Commands 
// Install a new tool
jarvis_server({ action: "install", name: "brave-search" })

// Switch project profile
jarvis_profile({ action: "edit", name: "my-project", add_servers: "context7,firecrawl" })

// Configure a client
jarvis_client({ action: "edit", client_name: "opencode", add_profile: "memory" })

// Check system health
jarvis_check_status()

// Bootstrap everything
jarvis_system({ action: "bootstrap" })

// Debug when tools fail to load
jarvis_diagnose({ action: "profile_health" })
jarvis_diagnose({ action: "logs", profile: "research" })

// Audit config synchronization
jarvis_diagnose({ action: "config_sync" })

Universal Compatibility

Models: Any MCP-compatible model — Claude, GPT, Gemini, DeepSeek, Llama

Clients: Claude Desktop, Claude CLI, Cursor, Windsurf, VS Code, Zed, OpenCode, Kilo Code

Client Configuration 
{
  "mcpServers": {
    "jarvis": {
      "command": "/path/to/Jarvis/jarvis",
      "args": []
    }
  }
}

That's it. Jarvis manages everything else.

Why Jarvis?

Without Jarvis With Jarvis
Edit JSON configs manually Agent installs tools via natural language
Restart client for new tools Hot-load tools mid-conversation
Duplicate configs per client One source of truth, multi-client
Fixed tool set per session Dynamic capabilities on-demand
Manual infrastructure repair Self-healing
Large tool definitions (~2,750 tokens) Consolidated tools (~1,350 tokens)

Roadmap

See ROADMAP.yaml for the full development plan.

Now

Feature Description
Daemon Stability Auto-restart failed servers with backoff, health checks per profile
Config Hot Reload Watch config files and apply changes without manual restarts
Smart Error Recovery Actionable recovery suggestions when tool calls fail

Next

Feature Description
Profile Templates Pre-built profiles for research, coding, data workflows
Multi-Model Optimization Context-aware tool pruning for different LLM providers
Usage Analytics Track tool usage patterns, error rates, response times

Proposed

Feature Description
Granular Client Editing Fine-grained server config (env vars, headers, enable/disable) via jarvis_client
Recently Completed
  • v5.1 — Docker operations, test runner, config backup/restore
  • v3.1 — jarvis_diagnose tool for self-debugging
  • v3.0 — Tool consolidation (24 → 9 tools, 52% token reduction)
  • HTTP API Transport — MCPM REST API with CLI fallback

Documentation

Doc Description
Examples Workflow examples
FAQ Common questions
Architecture Technical deep dive
Configuration Micro-Profile Strategy
Troubleshooting Issue resolution
API Reference Complete tool reference

Contributing

git clone https://github.com/YOUR_USERNAME/Jarvis-mcpm.git
cd Jarvis && go build -o jarvis . && go test -v ./...

MIT License · Issues · Discussions

About

Next-Gen MCP Gateway. Give your AI agent the ability to dynamically manage its own MCP tools.

Resources

Readme
Activity

Stars

2 stars

Watchers

1 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

No packages published

Contributors 3

  •  
  •  
  •  

Languages