Skip to main content
Glama
AAH20

Swarm-As-MCP

by AAH20
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Local

❖ Swarm-As-MCP

Universal Multi-Agent Swarm Exposer for Cursor, Windsurf, and Claude Code
Exposes entire autonomous multi-agent engineering swarms (Claude Opus 5.5, GPT-6 Astra, DeepSeek V4.1-Flash) as a single, zero-friction Model Context Protocol (MCP) server. Drop one line into your ~/.cursor/mcp.json to dispatch 10-agent fleets directly from your IDE editor.

License Python MCP Tests

⚡ The Problem: The Isolated Swarm Silo

Autonomous swarms are powerful, but developers live in their IDEs (Cursor, Windsurf, Claude Code):

  1. Tooling Fragmentation: Developers must leave their IDE, open a terminal, launch a multi-agent framework, wait, and manually copy code diffs back.

  2. Underpowered MCP Ecosystem: 99% of MCP servers are simple wrappers around SQL or single REST endpoints.

  3. No Streaming Multi-Agent Feedback: IDEs cannot observe what a swarm is doing in the background.

Swarm-As-MCP bridges the gap by wrapping an entire autonomous swarm into a single native MCP tool endpoint (@swarm_execute):

  • One-Click IDE Integration: Add one line to ~/.cursor/mcp.json.

  • Streaming Progress HUD: Emits real-time milestone events (planning, coding, compiling, virtual browser testing) straight into your IDE conversation.

  • Unified Atomic Output: Returns verified git diffs with 100% passed unit tests directly into your editor buffer.

Related MCP server: @willpowell8/cursor-cloud-agent-mcp

📐 Architecture & IDE Integration Flow

flowchart TD
    subgraph DeveloperIDE["Developer IDE (Cursor / Claude Code / Windsurf)"]
        User["Developer in Editor"]
        Prompt["Prompt: '@swarm refactor payment service and verify in UI'"]
        User --> Prompt
    end

    subgraph SwarmAsMCP["Swarm-As-MCP Gateway Server"]
        JSONRPC["Model Context Protocol (JSON-RPC 2.0)\nEndpoint: @swarm_execute"]
        Orchestrator["SwarmOrchestrator\n(Task Decomposition & Progress Streamer)"]

        Prompt --> JSONRPC
        JSONRPC --> Orchestrator
    end

    subgraph AutonomousFleet["Frontier Swarm Fleet"]
        Supervisor["Claude Opus 5.5 (Supervisor)"]
        CLIWorker["DeepSeek V4.1-Flash (Compiler & Bash)"]
        GUIWorker["GPT-6 Astra (Ghost Chromium Browser)"]

        Orchestrator --> Supervisor
        Supervisor --> CLIWorker
        Supervisor --> GUIWorker
    end

    subgraph StreamingResponse["Streaming Return to IDE"]
        Progress["Streaming Progress: 20% -> 50% -> 75% -> 100%"]
        GitDiff["Clean Git Diff & Test Results"]
        
        Orchestrator --> Progress
        Orchestrator --> GitDiff
        Progress --> User
        GitDiff --> User
    end

🚀 Quickstart

1. Installation

cd projects/swarm_as_mcp
pip install -e .

2. Configure Cursor / Windsurf (~/.cursor/mcp.json)

{
  "mcpServers": {
    "autonomous-swarm": {
      "command": "swarm-as-mcp",
      "args": ["serve"]
    }
  }
}

3. Run the IDE Simulation Demo

python3 -m swarm_as_mcp.cli demo

Output:

======================================================================
❖ SWARM-AS-MCP: UNIVERSAL MULTI-AGENT SWARM EXPOSER FOR IDEs
======================================================================
Frontier Swarm Fleet: Claude Opus 5.5 + DeepSeek V4.1-Flash + GPT-6 Astra
Target IDEs:          Cursor, Windsurf, Claude Code, OpenAI Operator
----------------------------------------------------------------------
 [████░░░░░░░░░░░░░░░░]  20% | Supervisor   (Claude Opus 5.5): Analyzing repository AST
 [██████████░░░░░░░░░░]  50% | CLI Worker   (DeepSeek V4.1-Flash): Refactored payment gateway
 [███████████████░░░░░]  75% | CLI Worker   (DeepSeek V4.1-Flash): Ran pytest suite: 42 passed
 [██████████████████░░]  90% | GUI Operator (GPT-6 Astra): Verified 200 OK checkout UI flow
 [████████████████████] 100% | Supervisor   (Claude Opus 5.5): Verified zero regressions
----------------------------------------------------------------------
[STEP 3] MCP TOOL RESULT RETURNED DIRECTLY TO IDE EDITOR:
 ✓ Execution Time: 0.12s
 ✓ Tests Passed:   42 tests

🧪 Testing

python3 -m unittest discover -s tests

Result: Ran 3 tests in 0.000s ... OK (100% passing)

📜 License

Apache-2.0. Copyright (c) 2026 AAH20.

This server cannot be deployed

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Maintenance

ActivityMaintained
ResponsivenessNo issues

Resources

Related MCP Connectors

Related MCP Servers

  • Coding Tools MCP

    xyTom
    A
    license
    Not graded
    quality
    B
    maintenance
    Empower any MCP-compatible AI Agent(MCP Client) with engineering-grade capabilities to understand, modify, run, and deliver real-world code repositories.
    348 PyPI
    1,155
    Apache 2.0

  • Codex Harness MCP

    2091k
    F
    license
    Not graded
    quality
    A
    maintenance
    Enables MCP-compatible AI clients to invoke CLI-driven agent tools over Streamable HTTP, including shell execution, file operations, patching, image viewing, web search, and nested agent tasks, with permission modes and real-time progress streaming.
    -
  • codeswarm by aaxnet

    codeswarmofficial

    aaxnet
    A
    license
    Not graded
    quality
    A
    maintenance
    Enables controlling an autonomous multi-agent engineering swarm from MCP clients, including starting and managing tasks, spawning agents, triggering PR reviews, and checking swarm status.
    5 npm
    MIT