VOXELL FORGE // MCP + OPENAI-COMPATIBLE

One embedding API. Every agent.

Give your coding agent production embeddings without building another integration. Forge MCP exposes embed and list_models through one stdio package. For workflow tools, swap one OpenAI base URL and keep the rest of your stack.

TURBO FREE · NO CARD · EMBEDDINGS ONLY · NO REQUEST DATA STORED

agent.config

> npx -y @voxell/forge-mcp

registry

name ai.voxell/forge

transport stdio

secret FORGE_API_KEY

tools

✓ embed

✓ list_models

01OFFICIAL MCP REGISTRY
02ONE REQUIRED SECRET
03THREE QUALITY TIERS
04OPENAI-COMPATIBLE API

Paste once. Give your agent embeddings.

Create a free Forge key, replace the placeholder, and paste the configuration into your MCP client. The package runs locally over stdio and sends embedding requests to Forge.

PACKAGE @voxell/forge-mcp

Public npm package. No global install required.

ENVIRONMENT FORGE_API_KEY

The only required secret. Get one from the dashboard.

TOOLS embed + list_models

Generate vectors or inspect available Forge models.

STANDARD MCP JSON Claude Desktop, Cursor, Windsurf, Gemini CLI
{
  "mcpServers": {
    "forge": {
      "command": "npx",
      "args": ["-y", "@voxell/forge-mcp"],
      "env": {
        "FORGE_API_KEY": "your-forge-api-key"
      }
    }
  }
}
VS CODE MCP.JSON .vscode/mcp.json or user profile
{
  "servers": {
    "forge": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@voxell/forge-mcp"],
      "env": {
        "FORGE_API_KEY": "your-forge-api-key"
      }
    }
  }
}
ONE-LINE INSTALLS Current CLI syntax for Claude Code, Codex, and Gemini CLI
CLAUDE CODE claude mcp add forge --env FORGE_API_KEY=your-forge-api-key -- npx -y @voxell/forge-mcp
CODEX codex mcp add forge --env FORGE_API_KEY=your-forge-api-key -- npx -y @voxell/forge-mcp
GEMINI CLI gemini mcp add forge npx -y @voxell/forge-mcp --env FORGE_API_KEY=your-forge-api-key

Use the client you already ship with.

Forge uses standard stdio MCP. Pick your platform, copy the matching block, and restart its MCP connection.

CC

Claude Code

Run the one-line install, then verify Forge with /mcp.

OX

Codex

Add Forge to the shared Codex MCP configuration, then run codex mcp list.

CU

Cursor

Paste the standard block into project .cursor/mcp.json or your user configuration.

WS

Windsurf

Paste the standard block into ~/.codeium/windsurf/mcp_config.json.

GC

Gemini CLI

Use the one-line install or add the standard block to settings.json.

VS

VS Code

Paste the VS Code block into .vscode/mcp.json or your user profile.

CD

Claude Desktop

Paste the standard block into the Claude Desktop MCP configuration and restart.

M+

Any stdio MCP client

Command npx, args -y @voxell/forge-mcp, env FORGE_API_KEY.

Set once, in any OpenAI-compatible tool.

These tools already accept an OpenAI-compatible embedding endpoint. Keep their native node or provider, set the Forge base URL, add your API key, and use the familiar model aliases.

BASE URL
https://api.voxell.ai/v1

Works with OpenAI-compatible embedding clients.

MODEL
text-embedding-3-small

text-embedding-3-large is also supported.

1. Choose the OpenAI embedding node 2. Paste the Forge base URL 3. Add your key

TOOL WHERE TO LOOK FIELD TO SET
n8n OpenAI Embeddings node Base URL
Flowise OpenAI Embeddings node Base Path
Langflow Embedding Model component API Base URL
Dify OpenAI-API-compatible provider
Settings → Model Provider → add
Endpoint URL
Open WebUI OpenAI engine
Admin Panel → Documents
API Base URL
LibreChat OpenAI-compatible endpoint provider Endpoint
AnythingLLM Generic OpenAI embedder Base URL

Dial quality, dimensions, and cost.

The same API spans three Forge tiers. Matryoshka dimensions let you shorten vectors and re-normalize them when storage matters.

TURBO 1024d

Fast and economical for high-volume embedding work.

PRO 2560d

More capacity for harder retrieval and semantic matching.

ULTRA 4096d

Highest-quality Forge tier for demanding embedding workloads.

Voxell's Ingot-8B-R3 research model ranks #1 for English on the public MTEB leaderboard, with a 75.98 mean across 41 tasks. This model receipt is separate from the callable Forge tiers.

Your agent already speaks the protocol.

Get a key, paste one block, and put production embeddings inside the builder path you use today.