Agent platform · Anthropic
Claude Code
Terminal-first agent with programmable hooks and a 1M-token window over the whole repo.
- MCP support
- native
- Where it runs
- terminal, ide, cloud, desktop
- Pricing
- Included in Claude paid plans; API usage billed separately
- Evidence grade C
- Mixed sourcing, or vendor figures with partial corroboration
Best for
- Large terminal-driven refactors across many files
- Repo-scale work without a prebuilt index
- Teams that want the agent loop scripted via hooks
Watch out
Reported coding scores travel with the model, not the harness - the same platform behaves differently per model and per scaffold. Benchmark it on your own repo before standardising.
In the FrankX stack
The default driver for FrankX agentic builds. The 30 hook events are what make it programmable rather than merely conversational.