Claude Code via Tokonomix

//What is Claude Code?

Claude Code is Anthropic's agentic coding tool: it reads your codebase, edits files across many directories, runs commands, and opens pull requests. Unlike most CLI agents it spans surfaces that share one engine — a Terminal CLI, a VS Code extension (inline diffs, @-mentions, plan review), a JetBrains plugin, a Desktop app, and the web at claude.ai/code. Your CLAUDE.md files, settings and MCP servers carry across all of them.

Most surfaces use a Claude subscription or Anthropic Console account — but the Terminal CLI and VS Code also support third-party providers. That hook is where Tokonomix plugs in: set the Tokonomix base URL plus your API key and Claude Code’s calls route through a single key with access to 132 models from Anthropic, OpenAI, Google, Mistral, Meta and others.

As an agent, Claude Code does real work rather than just suggesting completions: it plans a change, edits across multiple files, runs your tests, fixes what breaks, and stages a commit or opens a pull request. It can spawn sub-agents for parallel workstreams, run shell hooks before and after edits (auto-format, lint-on-commit), and read project rules you keep in a CLAUDE.md file at the repo root. Because the same engine drives every surface, a workflow you set up in the terminal behaves identically inside VS Code or a JetBrains IDE — and so does the Tokonomix integration described below.

//Why route Claude Code through Tokonomix?

Claude Code is excellent at Anthropic’s own models, but a single model — however good — has blind spots, and not every task wants the same engine. Routing through Tokonomix gives you a model layer that is broader (every major family behind one key), safer on the decisions that matter (a consensus verdict instead of one opinion), and fully accountable (per-call cost and region control). You keep Claude Code’s agentic workflow; Tokonomix changes only what sits behind the API call.

• Cross-family consensus on demand. Add the Tokonomix MCP server and Claude Code can call tokonomix_consensus_ask on any high-stakes refactor or design decision — 2–6 frontier models cross-checked by a judge, so a confident-but-wrong step gets caught before it lands in your repo.
• One key, 132 models. Reach non-Anthropic models through the same key without separate accounts.
• Sovereignty routing — your choice, not EU-only. Pin models by where they run (hosting_region) or where they're made (origin_country) — EU-only, US-only, a single country, or a custom set. Enforce it per call, or set a predefined rule (e.g. auto-select EU models) in your dashboard. EU/GDPR is just one example.
• Transparent per-call cost. Every call is metered to the cent in your Tokonomix dashboard, so an agent that fires hundreds of calls a day never surprises you.
• Provenance & sovereignty. Filter by origin_country to pick models by where they were built — not just where they run — when procurement or sovereignty rules demand it.

//Install Claude Code — Windows, macOS, Linux (also WSL)

Native installers auto-update in the background. Git for Windows is recommended on native Windows so Claude Code can use its Bash tool.

irm https://claude.ai/install.ps1 | iex

curl -fsSL https://claude.ai/install.sh | bash
# or
brew install --cask claude-code

curl -fsSL https://claude.ai/install.sh | bash

Then start it in any project:

cd your-project
claude

//Enter your Tokonomix key

Already have a key? Claude Code reads a third-party gateway from two environment variables. Generate a key (tok_live_…) in your Tokonomix dashboard and point Claude Code at the Anthropic-compatible endpoint:

# route Claude Code through Tokonomix
export ANTHROPIC_BASE_URL=https://tokonomix.ai/api/anthropic
export ANTHROPIC_API_KEY=tok_live_your-key
claude

//Pull in skills & MCP tools

Two complementary layers — base-URL passthrough for everyday model access, and MCP for consensus on demand.

# already wired by ANTHROPIC_BASE_URL
claude "refactor the auth module"

claude mcp add tokonomix -- \
  npx -y tokonomix-mcp
# set the key for the server
TOKONOMIX_API_KEY=tok_live_your-key

The Tokonomix MCP server exposes tokonomix_consensus_ask, tokonomix_single_ask, tokonomix_list_models and tokonomix_get_skill. Claude Code will call consensus when a task is high-stakes; you can also nudge it explicitly ("get a consensus verdict on this migration"). The two routes compose well: keep base-URL passthrough for the fast, high-volume edits Claude Code makes all day, and let it reach for the consensus tool only at the moments where being wrong is expensive — a schema migration, an auth change, a public API contract. Browse Claude Code’s own skills with /skills, package repeatable workflows as custom skills your team shares, and keep coding standards and architecture notes in CLAUDE.md so every session — on any surface — starts with the same context.

//Using Claude Code inside an IDE

Here Claude Code differs from terminal-only agents: it ships native editor extensions, so the agent lives inside your IDE with inline diffs — not just in a terminal pane. The Tokonomix env vars and MCP server apply identically across every surface, because they all share one engine.

//Verify it works

After exporting the env vars, run claude and ask for a small reversible change — say, add a docstring. You should see a metered call in your Tokonomix dashboard. Then add the MCP server and ask Claude Code to "get a consensus verdict" on a risky change; the multi-model result comes back as the MCP tool result.