Skip to main content
Glama
Sign In
fusion-mcp

fusion-mcp

by yksanjo
OverviewSchemaSourceRelated ServersDiscussionsAdmin
JavaScript
Remote

fusion-mcp

A zero-dependency MCP server that exposes OpenRouter Fusion as a tool any agent can call.

Fusion sends your prompt to a panel of models in parallel (with web search), a judge model compares their answers, and you get back one synthesized answer — stronger than any single model. This wraps it so Claude Code (or any MCP client) can reach Fusion directly instead of going through the OpenRouter web UI.

Tool

fusion_ask

arg

required

description

prompt

✅

The question or task for the Fusion panel.

analysis_models

1–8 model slugs forming the panel. Omit for Fusion's default preset.

judge_model

Model that synthesizes the panel's answers. Defaults to the first panel model.

preset

Curated preset: general-high or general-budget.

max_tool_calls

Tool iterations per panel model (1–16, default 8).

Fusion runs several completions per call (panel + judge), so it costs more than a single model call. Use it where being wrong is costly: research, critique, decisions.

Related MCP server: mcp-fusion

Setup

Requires Node ≥ 18 and an OpenRouter API key.

export OPENROUTER_API_KEY=sk-or-...

Wire into Claude Code

claude mcp add fusion -e OPENROUTER_API_KEY=sk-or-... -- node /Users/yoshikondo/fusion-mcp/server.mjs

Or add to your MCP config manually:

{
  "mcpServers": {
    "fusion": {
      "command": "node",
      "args": ["/Users/yoshikondo/fusion-mcp/server.mjs"],
      "env": { "OPENROUTER_API_KEY": "sk-or-..." }
    }
  }
}

Then in a session: "use fusion_ask to research X".

Test

npm test

The test suite spins up a real local HTTP server standing in for OpenRouter and drives a real child-process MCP server over real stdio — nothing internal is mocked.

License

MIT

Available Tools

1 tool
fusion_askA

Ask OpenRouter Fusion: a panel of models answers your prompt in parallel (with web search), a judge compares them, and one synthesized answer is returned — stronger than any single model. Use for research, expert critique, or any question where being wrong is costly. Costs more than a single call (panel + judge completions).

Inputs: prompt (required), preset, judge_model, max_tool_calls, +1 more.

Tool Schema Changelog

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

Loading tool schema changelog…

    Deploy Server
    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

    TDQS

    A4.3/5.0

    Scored across 1 tool

    Disambiguation5/5

    With only one tool, there is no possibility of overlapping or confusing tools. The single tool's purpose is clearly defined and distinct.

    Naming Consistency5/5

    With a single tool, the naming is trivially consistent. The verb_noun pattern (fusion_ask) is clear and well-formed.

    Tool Count2/5

    A single tool is too few for a typical MCP server, making the tool surface feel sparse. Even if the tool is powerful, the server lacks the breadth expected of a useful toolkit.

    Completeness4/5

    The server fully covers its stated purpose of asking a fusion panel, but there are minor gaps such as no configuration options or alternative request modes. The core functionality is complete, but some auxiliary features are missing.

    Maintenance

    ActivityInactive
    ResponsivenessOutdated data

    Resources

    Related MCP Connectors

    Related MCP Servers