Skip to main content
Glama

godmod3-mcp

MCP bridge for G0DM0D3 with dedicated skills for Odysseus and Hermes Agent.

This package delegates 100% to a self-hosted G0DM0D3 API and exposes every flagship feature as an MCP tool:

  • ULTRAPLINIAN multi-model racing

  • CONSORTIUM hive-mind synthesis

  • Single-chat with GODMODE / AutoTune / Parseltongue / STM

  • AutoTune analysis, Parseltongue obfuscation, STM transforms

  • Dataset + research endpoints

  • Local OpenAI-compatible model support (Ollama, LM Studio, vLLM)

Quick start

Run the upstream G0DM0D3 API as a container, run a local stdio bridge for Hermes, and run a separate HTTP bridge container for Odysseus.

1. Start the G0DM0D3 API container

cp .env.example .env
# edit .env with optional keys
docker compose -f docker-compose.api.yml up -d --build

The API is available at http://localhost:7860 from the host and at http://godmod3-api:7860 from containers on the shared godmod3 network.

Note: The upstream G0DM0D3 API currently has a path-to-regexp compatibility issue (/batch/* route pattern). The local Dockerfile.api clones upstream and applies a small patch before building so the container starts correctly.

Avoid upstream rate limits

By default the API runs in Free tier mode: 5 total requests, 10/min, 50/day. To get unlimited local use, set a tier key in .env:

GODMODE_API_KEY=my-local-key-123
GODMODE_TIER_KEYS=enterprise:my-local-key-123

Then pass the same key to the bridge via GODMOD3_API_KEY. Restart the API container to apply.

2. Install the bridge locally for Hermes (stdio)

pip install -e .

This installs the bridge and the console script godmod3-mcp.

3. Start the HTTP bridge container for Odysseus

docker compose -f docker-compose.bridge-http.yml up -d --build

Odysseus connects to http://localhost:3001/sse.

4. Verify

python -m godmod3_mcp.server --test

Alternative: run the API directly

If you prefer not to containerize the API:

git clone https://github.com/elder-plinius/G0DM0D3.git
cd G0DM0D3
npm install
npm run api

Alternative: full stack at once

docker compose up -d --build

This is equivalent to starting both docker-compose.api.yml and docker-compose.bridge-http.yml together.

Related MCP server: Multi-CLI MCP

Configuration

Variable

Default

Description

GODMOD3_BASE_URL

http://localhost:7860

G0DM0D3 API endpoint

GODMOD3_API_KEY

(none)

Bearer token if API requires auth

GODMOD3_MCP_TRANSPORT

stdio

stdio, http, or sse

GODMOD3_MCP_HTTP_PORT

3001

HTTP/SSE listen port

GODMOD3_MCP_ALLOWED_HOSTS

(none)

Comma-separated allowed HTTP Host header patterns for the HTTP/SSE transport (e.g. godmod3-mcp-http:*)

GODMOD3_MCP_DISABLE_DNS_REBINDING_PROTECTION

false

Set to true to disable MCP SDK Host-header validation (only in trusted networks)

MCP Tools

  • health_check

  • server_info

  • list_models

  • get_tier

  • single_chat

  • ultraplinian_chat

  • consortium_chat

  • autotune_analyze

  • parseltongue_encode

  • parseltongue_detect

  • transform_text

  • submit_feedback

  • dataset_stats

  • export_dataset

  • research_info

  • research_stats

  • research_query

Connect to Odysseus

Via the MCP admin UI or API:

{
  "name": "godmod3",
  "transport": "stdio",
  "command": "python",
  "args": ["-m", "godmod3_mcp.server"],
  "env": {
    "GODMOD3_BASE_URL": "http://localhost:7860",
    "GODMOD3_API_KEY": "optional-key"
  }
}

Or via the HTTP bridge container:

{
  "name": "godmod3",
  "transport": "http",
  "url": "http://localhost:3001/sse"
}

Install the skill:

cp -r skills/odysseus/godmod3 /path/to/odysseus/data/skills/

Connect to Hermes Agent

Hermes launches the bridge as a local stdio subprocess. Make sure you installed the bridge with pip install -e . and that the API container is exposing port 7860.

Add to your Hermes MCP config (file path depends on install; often ~/.hermes/mcp_servers.json or via hermes config):

{
  "godmod3": {
    "command": "godmod3-mcp",
    "env": {
      "GODMOD3_BASE_URL": "http://localhost:7860",
      "GODMOD3_API_KEY": "optional-key"
    }
  }
}

Install the skill:

cp -r skills/hermes/godmod3 ~/.hermes/skills/

Then invoke with /godmod3.

Connect other MCP clients (Claude Code, Codex, Cursor, Kimi Code)

The bridge is a standard MCP server. Any MCP-compatible client can connect via stdio or HTTP/SSE.

{
  "mcpServers": {
    "godmod3": {
      "url": "http://localhost:3001/sse"
    }
  }
}

stdio

{
  "mcpServers": {
    "godmod3": {
      "command": "/home/vudu/.venv/godmod3-mcp/bin/godmod3-mcp",
      "env": {
        "GODMOD3_BASE_URL": "http://localhost:7860",
        "GODMOD3_API_KEY": "optional-key"
      }
    }
  }
}

Client-specific skill files

Copy the appropriate skill file into your client config:

Client

File

Typical location

Claude Code

skills/claude-code/CLAUDE.md

Project root or ~/.claude/CLAUDE.md

Cursor

skills/cursor/.cursorrules

Project root

Codex

skills/codex/CODEX.md

Project root or Codex config

Generic / Kimi Code

skills/generic-mcp/SKILL.md

Use as system prompt or skill

These files explain when and how to use the G0DM0D3 tools in each client.

Local model support

Pass local_model_url and local_models to any chat tool:

{
  "messages": [{"role": "user", "content": "Hello"}],
  "local_model_url": "http://localhost:11434/v1",
  "local_models": ["qwen3:8b"],
  "provider_preference": "all"
}

Tests

pytest tests/test_tools.py

License

AGPL-3.0-or-later — same as G0DM0D3.

Available Tools

17 tools
autotune_analyzeC

Analyze a message and get optimal LLM sampling parameters from AutoTune.

Inputs: message (required), strategy, conversation_history. Returns: result

consortium_chatC

Collect responses from all tier models and synthesize ground truth (CONSORTIUM).

Inputs: messages (required), tier, stream, godmode, +13 more. Returns: result

dataset_statsB

Show in-memory dataset statistics.

Inputs: no named parameters. Returns: result

export_datasetA

Export the dataset as JSON or JSONL.

Inputs: format. Returns: result

get_tierA

Show your current G0DM0D3 API tier, limits, and features.

Inputs: no named parameters. Returns: result

health_checkA

Check connectivity to the G0DM0D3 API.

Inputs: no named parameters. Returns: result

list_modelsA

List available models including virtual ULTRAPLINIAN/CONSORTIUM models.

Inputs: no named parameters. Returns: result

parseltongue_detectC

Detect trigger words without transforming them.

Inputs: text (required), custom_triggers. Returns: result

parseltongue_encodeC

Obfuscate trigger words in text.

Inputs: text (required), intensity, technique, custom_triggers. Returns: result

research_infoB

Return research dataset schema and repository info.

Inputs: no named parameters. Returns: result

research_queryC

Query the published research corpus with filters.

Inputs: mode, limit, model, since, +3 more. Returns: result

research_statsB

Return aggregate stats across published HuggingFace batches.

Inputs: no named parameters. Returns: result

server_infoA

Return G0DM0D3 API info, endpoints, and tier defaults.

Inputs: no named parameters. Returns: result

single_chatC

Single-model chat with the full G0DM0D3 pipeline (GODMODE, AutoTune, Parseltongue, STM).

Inputs: messages (required), model, top_k, top_p, +19 more. Returns: result

submit_feedbackC

Submit quality feedback for the AutoTune EMA learning loop.

Inputs: rating (required), message_id (required), params, context_type, +1 more. Returns: result

transform_textC

Apply semantic transformation modules (STM) to text.

Inputs: text (required), modules. Returns: result

ultraplinian_chatC

Race many models in parallel and return the best response (ULTRAPLINIAN).

Inputs: messages (required), tier, stream, godmode, +15 more. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3/5.0

    Scored across 17 tools

    Disambiguation4/5

    Most tools have clearly distinct purposes, such as single_chat vs. ultraplinian_chat vs. consortium_chat, and parseltongue_encode vs. parseltongue_detect. Minor overlap exists between health_check and server_info (both relate to connectivity/info) and between dataset_stats and research_stats (both provide statistics but on different datasets), but these are distinguishable with careful reading.

    Naming Consistency3/5

    Naming conventions are mixed: some tools follow verb_noun patterns (list_models, get_tier, transform_text, submit_feedback, export_dataset), while others use noun_verb or compound nouns (health_check, parseltongue_encode, parseltongue_detect, server_info, dataset_stats). All names are readable and use snake_case, but the inconsistent verb placement and noun-only names prevent a higher score.

    Tool Count3/5

    With 17 tools, the server falls into the 'heavy' range (16-25). The tool count is justified by the broad scope of the API, covering chat, optimization, text transformation, and research, but it still feels slightly bloated and could be consolidated by merging some informational tools or research stats endpoints.

    Completeness4/5

    The tool surface is largely complete for the stated domain: it covers chat (single, ULTRAPLINIAN, CONSORTIUM), AutoTune analysis, Parseltongue encoding/detection, STM transformation, feedback, dataset export, and research querying. Minor gaps include no Parseltongue decoding and no dataset upload/management, but these are not critical for the core API usage.

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      D
      maintenance
      Bridges Google's Gemini CLI with MCP-compatible clients and integrates OpenRouter API for access to 400+ AI models. Provides 33 specialized tools enabling multi-AI workflows, collaborations, and debates between Claude, Gemini, and other LLMs.
      142
      -
    • A
      license
      Not graded
      quality
      A
      maintenance
      Multi-server MCP tool bridge that exposes banking, email, calendar, and custom API tools for AI clients via the Model Context Protocol.
      3,833 npm
      MIT