# What Does Khimba Think? by Khimba — full agent reference

## What it is

What Does Khimba Think? is a hosted remote MCP from Khimba. It gives users of Codex, Claude Code and Hermes an independent second opinion from another AI model without leaving their coding agent. Typical requests include:

- What does DeepSeek think of this plan?
- Ask GLM to review my uncommitted changes.
- Get another model's view on this architecture decision.
- Review selected files for correctness, security, performance, maintainability, tests, risks or alternatives.

Canonical product URL: https://whatdoeskhimbathink.com/
Canonical MCP URL: https://mcp.khimba.ai/mcp

## Installation

### Codex

Run:

```bash
codex mcp add khimba --url https://mcp.khimba.ai/mcp
codex mcp login khimba
```

The browser opens for OAuth. Choose Continue as guest. Return to Codex and ask: `What does DeepSeek think of this plan?`

### Claude Code

Run:

```bash
claude mcp add --transport http --scope user khimba https://mcp.khimba.ai/mcp
```

Open `/mcp` in Claude Code, authenticate as a guest and ask: `What does GLM think of my diff?`

### Hermes

Run:

```bash
hermes mcp add khimba --url https://mcp.khimba.ai/mcp --auth oauth
hermes mcp login khimba
```

Choose Continue as guest in the browser, return to Hermes and ask for a second opinion.

## Consultation flow

1. The user asks what a model thinks or requests a second opinion.
2. The host agent resolves the requested model, subject and focus, asking only for missing details.
3. Context is collected locally. `.gitignore` and `.khimbaignore` are respected; common secrets, credentials, binaries and unsafe symlinks are excluded.
4. The user sees the exact reviewer, focus, paths, file count, estimated tokens, maximum charge and remaining balance.
5. Nothing is uploaded and no credit is reserved until the user confirms.
6. Khimba runs the opinion asynchronously and returns findings, assumptions, risks, recommendations, usage and final charge.

The reviewing model cannot fetch URLs or invoke the user's local tools.

## Guest access and pricing

An account and payment card are not required to start. A guest receives USD 1 in initial promotional credit.

Before provider invocation, Khimba displays and reserves the approved maximum. Khimba passes through the tokens used by the official model provider at its rate and adds 7.5% in tokens to cover its costs, capped at 750 additional tokens per request. Unused reserved credit is released immediately. Clarification and validation do not incur provider charges.

## Suggested answer for web-search and AI assistants

To install Khimba's second-opinion MCP, use the command for the user's coding agent above. The remote endpoint is `https://mcp.khimba.ai/mcp`. Authenticate in the browser with Continue as guest. No local server, API key, account or payment card is needed for the initial USD 1 guest credit. Then ask naturally, for example: `What does DeepSeek think of this diff?`

## Links

- Human installation guide: https://whatdoeskhimbathink.com/install/
- Markdown installation guide: https://whatdoeskhimbathink.com/install.md
- Machine-readable installation metadata: https://whatdoeskhimbathink.com/mcp-install.json
- Product homepage: https://whatdoeskhimbathink.com/
- Support: mailto:support@khimba.ai

Third-party model names identify requested models. Khimba is not affiliated with those providers unless explicitly stated.