A dedicated contentrain branch holds content state, each write forks a cr/ branch in a temporary worktree, and the result merges back with canonical JSON formatting. Nothing stashes or checks out under you while you work. The tool list is also capability-aware: a session driven by a remote provider with no local checkout only advertises the tools it can actually satisfy, rather than failing halfway through one.
The engine behind Contentrain's MCP integration. It manages a .contentrain/ directory in your project — models, content, locales — and every write goes through a git transaction rather than straight onto disk.
- Project status, a model's full schema with sample data, and the file-format contract itself — contentrain_status, contentrain_describe, contentrain_describe_format
- A project initialized from scratch or from a starter template — blog, landing, docs, ecommerce, saas, i18n, mobile — contentrain_init, contentrain_scaffold
- Models created, updated and deleted, with content and meta files following — contentrain_model_save, contentrain_model_delete
- Content entries saved, listed with filters and locales resolved, and deleted down to individual dictionary keys — contentrain_content_save, contentrain_content_list, contentrain_content_delete
- Validation against the model schemas — required fields, type mismatches, broken relations, secret leaks, i18n parity — with structural problems auto-fixed on request — contentrain_validate
- Source code scanned for hard-coded strings, then those strings extracted into content files or the source patched to reference them, with a dry run by default — contentrain_scan, contentrain_apply
- Batch operations across entries, including locale-to-locale work — contentrain_bulk
- Review flow in git: branches pushed for review, merged back into the contentrain branch locally, listed with their merge status, and deleted when abandoned — contentrain_submit, contentrain_merge, contentrain_branch_list, contentrain_branch_delete
- A health report covering git, node, structure, model parsing, orphaned content, branch pressure and SDK freshness — contentrain_doctor
Node.js 22 or newer and git on the machine; install @contentrain/mcp and run contentrain-mcp over stdio, with CONTENTRAIN_PROJECT_ROOT pointing at the project if it is not the working directory. It is local-first, with optional GitHub and GitLab backends and an HTTP transport for remote drivers such as Contentrain Studio. Optional parsers — @vue/compiler-sfc, @astrojs/compiler and svelte — sharpen source scanning for those frameworks; the scanner works without them.
One command — pnpm add @contentrain/mcp