Mimiran is a standalone MCP server that turns your repository into a self-verifying knowledge graph — machine-readable, consistent, and safe for AI agents to reason over.
# AI agent queries Mimiran for route context
→ CallTool context_by_route "/api/v1/payments"
# Response includes verified context + meta
← {
"route": "/api/v1/payments",
"feature": "billing",
"schema_hash": "c4a9f...",
"sync_state": "✓ fresh"
}Modern repositories inevitably drift. Docs fall behind code. Schemas go stale. Feature registries lie. And AI agents — making decisions over this — amplify every inconsistency. Manual consistency maintenance becomes impossible at scale.
Stop dealing with hallucinated APIs and outdated file references. Mimiran gives your AI assistant a structured, verified knowledge graph of your actual codebase.
Multiple developers using AI on the same repo? Everyone connects to one Mimiran project and gets identical, up-to-date context — no more diverging mental models.
When the entire project doesn't fit in an AI context window, Mimiran serves the right fragment on demand — routes, features, schemas — without loading everything.
Mimiran acts as a Context Controller for your engineering environment. It turns the repository into a self-verifying system of knowledge.
Mimiran is infrastructure — not a product feature. It plugs into your existing stack and makes it smarter.
Mimiran indexes your entire codebase and turns it into a structured, machine-readable knowledge graph. AI agents query context directly — no more reading dozens of files.
Keeps context/*, docs/*, MONOREPO_MAP.md and FEATURE_REGISTRY.md in sync with actual code. Supports full sync and incremental delta-sync sessions for active AI coding flows.
Enforce rules like no any, no console.log, cross-import restrictions, platform isolation — scanned across all .ts/.tsx files via CLI or the get_invariants MCP tool.
Parses Tables and Enums from types.ts, maps them to SQL migrations, tracks schema-impacting changes, and feeds structured DB context to agents.
context-cli.js ships with check, sync-all, sync-pages, sync-dependencies. Run the same validation locally, in pre-commit hooks, and on CI.
A single binary can serve multiple projects. Connect different repositories without reinstalling — one infrastructure layer for your whole org.
Mimiran exposes structured Read and Write tools over MCP. AI agents call them directly — no file parsing, no hallucinated context, no outdated snapshots.
get_tech_stackcontext_by_routecontext_by_featuresearch_contextget_feature_registryget_invariantsget_ai_rulesdb_describe_tabledb_list_tablesdb_list_enumssync_contextsync_monorepo_mapscaffold_feature_specscaffold_page_docopen_sync_sessionsync_session_batch+ more — 40 MCP tools total (21 read · 19 write)
The context-cli.js layer lets you run the same validation locally, in pre-commit hooks, and on CI — with identical results every time.
context-cli checkcontext-cli sync-allcontext-cli sync-pagescontext-cli sync-dependenciescontext-cli sync-invariantscontext-cli scaffold-*✓ Indexing repository structure...
✓ Validating page docs ↔ routes
✓ Checking markdown structure
✓ Verifying required contracts
✓ Scanning invariants (204 rules)
✓ Checking schema sync state
✓ Key artifacts synced
All checks passed. (0 errors, 0 warnings)ReadResource is protected against path traversal attacks. Files are served only from allowed directories, never outside project context.
All MCP calls are logged with secret masking and prompt/query truncation. Sensitive data never appears in plain logs.
Readonly mode blocks all Write tools, enabling safe audit runs without any risk of accidental state mutations.
What is shipped today is separated from roadmap packaging so the release page stays honest.
The only active plan during the testing period.
This tier is locked until the product release.
Enterprise packaging will only be offered after release.
Your project guarantees the correctness of its own context. Less architectural drift. Safe AI agents. Synchronized documentation.