// version intelligence for llms

YOUR LLM
GENERATES
STALE CODE

Claude generates Pydantic v1 patterns in a v2 project. GPT writes Terraform v4 S3 config when you're on v6. By the time terraform validate catches it, the wrong code is anchoring the context window.

CodeAdvisor front-loads version-accurate schemas into the LLM's context before generation begins.

How It Works

Connect via MCP. Claude Code queries CodeAdvisor automatically before generating code — checking schemas, validating attributes, and reading landmines.

$ validate_resource("pydantic", "pydantic.BaseModel",
    ["model_dump", "dict"], version="1.10.19")
 
## Valid (1)
  - dict
 
## Unknown / Likely Hallucinated (1)
  - model_dump
 
// model_dump is v2-only. Caught before generation.

MCP Tools

9 tools available when connected via MCP:

schema_lookup
Exact schema for a resource — attributes, types, deprecations
schema_search
Full-text search when you don't know the exact name
validate_resource
Check attributes: valid, deprecated, or hallucinated
recommend_version
Version recommendation with deltas from what the LLM "knows"
list_resources
List indexed resources with prefix filter
guidance_search
Semantic search over changelogs and migration notes
get_landmines
Known LLM failure patterns with correct alternatives
stack_check
Check multiple technologies at once
codeadvisor_rules
Behavioral prompt — tells the LLM to query before generating