mcp-second-opinion
Allows querying Google models (e.g., Gemini 2.5 Flash) to provide a second opinion on a prompt, as part of a multi-provider panel.
Allows querying Meta models (e.g., Llama) via OpenRouter to provide a second opinion on a prompt, as part of a multi-provider panel.
Allows querying OpenAI models (e.g., GPT-4o-mini) to provide a second opinion on a prompt, as part of a multi-provider panel.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@mcp-second-opinionIs this SQL query vulnerable to injection? Give me a second opinion."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
mcp-second-opinion
An MCP server that gets you a second opinion. It asks the same question to a panel of LLMs from different providers — in parallel, through a single OpenRouter key — and hands back every answer side by side. Optionally it synthesizes them, and is told to surface where the models disagree instead of averaging it into a confident-sounding blur.
One model can be confidently wrong. A panel that's forced to disagree is much harder to fool — so give your agent (Claude Desktop, or any MCP client) the ability to check a claim against three minds before trusting one.
You → second_opinion("Is this SQL injection-safe?", synthesize: true)
## openai/gpt-4o-mini → "Yes, parameterised — safe."
## anthropic/claude-3.5-haiku → "No — the ORDER BY clause is interpolated."
## google/gemini-2.5-flash → "Safe for the WHERE clause; ORDER BY is a risk."
## 🔎 Synthesis
Disagreement: the ORDER BY interpolation. Two of three flag it. Bottom line: not safe.Why
Cross-provider by design. OpenRouter proxies OpenAI / Anthropic / Google / Meta / … behind one OpenAI-compatible endpoint, so a real multi-provider panel needs only one API key.
Resilient by construction. Members run with
Promise.allSettledand a per-model timeout. A slow or failing model is reported as one failed answer — it never sinks the batch. (Covered by tests.)Disagreement is the point. The synthesizer is explicitly instructed to name conflicts, not smooth them away.
Related MCP server: Multi-Model Advisor
Install
Requires Node ≥ 18 and an OpenRouter API key (free to create at https://openrouter.ai/keys).
git clone https://github.com/aleksbuss/mcp-second-opinion.git
cd mcp-second-opinion
npm install
npm run buildAdd it to Claude Desktop
Edit claude_desktop_config.json (Claude → Settings → Developer → Edit Config):
{
"mcpServers": {
"second-opinion": {
"command": "node",
"args": ["/absolute/path/to/mcp-second-opinion/dist/index.js"],
"env": {
"OPENROUTER_API_KEY": "sk-or-..."
}
}
}
}Restart Claude Desktop; the second_opinion and list_panel_models tools appear.
Tools
second_opinion
arg | type | notes |
| string (required) | The question to put to the panel. |
| string[] (optional) | OpenRouter model ids. Omit to use the default panel. |
| boolean (optional) | If true, a synthesizer model compares the answers. |
| string (optional) | System prompt sent to every model. |
list_panel_models
Returns the configured default panel and how to override it. No arguments.
Configuration (environment)
var | required | default |
| ✅ | — |
|
| |
|
| |
|
|
Any model id on https://openrouter.ai/models works. Pick a cheap, fast panel — you're paying for N calls per question.
Develop
npm run typecheck # tsc --noEmit
npm test # vitest (mocks fetch; no network, no key needed)
npm run build # → dist/
npm run dev # run from source via tsxThe core fan-out and synthesis logic lives in src/panel.ts and is pure + fetch-injectable, so the resilience behaviour (one member failing, timeouts) is unit-tested without touching the network.
License
MIT © Aleksejs Buss
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Latest Blog Posts
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/aleksbuss/mcp-second-opinion'
If you have feedback or need assistance with the MCP directory API, please join our Discord server