Khimba/ Claude Code

Claude Code integration

Bring a different model into Claude Code.

Use Khimba to ask another model to challenge a diff, plan or architecture decision without manually moving repository context between applications.

Plugin: primary after approval

The What Does Khimba Think? plugin combines the hosted MCP tools with a local skill that collects repository evidence and enforces the confirmation workflow. It is the intended primary Claude Code route only after Anthropic approves and lists it. Khimba does not currently claim that approval or publish an unverified marketplace command.

Connector: all supported Claude surfaces

The remote Khimba connector is intended for Claude web, Desktop, mobile, Cowork, Claude Code and the API where those surfaces support connectors. It provides the MCP tools across surfaces but does not add the Claude Code plugin's local skill. Connector directory availability is also subject to Anthropic approval.

Direct MCP: tools-only fallback

claude mcp add --transport http --scope user khimba https://mcp.khimba.ai/tools/model-opinions

Open /mcp. Choose Sign in to Khimba for a permanent account with stable reconnect, or Continue as guest for temporary access. This command connects the remote tools only; use a host-supported skill installer separately if you need the local workflow.

Ask naturally

Try “What does GLM think of this diff?” or “Ask DeepSeek to challenge this migration plan and identify rollback risks.” With the plugin skill, the host collects context locally and presents a confirmation rather than immediately uploading the repository.

Inspect the request

Before inference, review the selected model, purpose, evidence paths, estimated tokens, maximum charge and remaining balance. A changed scope or model requires a new confirmation. Common secrets, credentials, binaries and unsafe symlinks are excluded, but no filter is perfect—inspect the manifest.

Use the answer responsibly

A different model may catch an omission because it approaches the work with different assumptions. It may also be confidently wrong. Verify important findings against source, tests and documentation. Model agreement is useful evidence, not proof.

Temporary guests can use up to USD 0.10 in promotional credit, limited to three successful calls and 24 hours after guest creation. The promotion ends when any one limit is reached. The final charge uses measured official-provider token cost plus a 30% Khimba platform markup.