MCP server for chameleon. Auto-derives codebase conventions and serves archetype-keyed guidance via stdio JSON-RPC.
chameleon-mcp has 12 well-defined tools with explicit schemas and descriptions. Tool naming follows verb_noun patterns consistently (e.g., chameleon-status, bootstrap-repo, teach-rule, list-rules, get-rules, extract-code). Descriptions are present and substantive (ranging 100-300+ chars), exceeding the 10-char minimum and falling within the productive 10-1024 range. All tools have explicit JSON Schema input definitions with type and description fields for parameters. However, there are gaps: (1) output schemas are not documented in the tool definitions, responses are described narratively but not as structured JSON Schema, (2) parameter descriptions could be more prescriptive about constraints (e.g., enum values, ranges, formats), (3) error handling guidance is absent, tools do not describe what happens on failure or recovery paths, (4) idempotency and side-effects are not explicitly declared, making retry semantics unclear for destructive operations like bootstrap-repo and teach-rule. The server is well-above-average for community MCP servers but falls short of production-grade (80+) due to missing output schemas, incomplete parameter constraint documentation, and lack of error recovery guidance.
Bootstrap a repository by deriving codebase conventions from its structure, tool configs, and code patterns. Creates or updates the .chameleon directory with rules.json and conventions.md.
Disable chameleon for the current session, preventing rule enforcement and guidance injection.
Enable chameleon for the current session, allowing rule enforcement and guidance injection.
Retrieve the status and configuration of the chameleon server, including bootstrap state, tool configs, rules, and enforcement settings.
Extract and parse code structure from source files using tree-sitter, returning AST information or function signatures.
Retrieve enforcement.json configuration for a repository, specifying which rules are actively enforced.
Retrieve idiom-keyed guidance for a repository, returning archetype-specific recommendations for code patterns.
Output schemas not documented. Tool definitions describe what is returned narratively ('returns AST information', 'returns archetype-specific recommendations') but do not specify structured JSON Schema for responses. LLMs cannot reliably parse unstructured output or plan downstream calls without knowing response field names and types.
Parameter constraints under-specified. Parameters like rule_value (object), enforcement_config (object), and extract_type (enum) lack detailed format descriptions. rule_value accepts 'The rule value or configuration object' but does not specify required/optional sub-fields or structure. extract_type enum is present in schema but description does not list the allowed values inline for easy LLM reference.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
Retrieve the complete rules.json configuration for a repository, with optional sanitization.
List all active rules for a repository, with optional filtering by rule type or source.
Refresh an already-bootstrapped repository, re-deriving conventions and updating rules.json without full re-bootstrap.
Update enforcement.json to activate or deactivate specific rules for a repository.
Teach chameleon a new rule or override an existing rule for a specific repository, updating rules.json and enforcement.json.
No error handling guidance. Tools do not document failure modes or recovery paths. E.g., bootstrap-repo does not specify: what happens if the repo path is invalid? What if .chameleon already exists? Should the LLM retry, ask the user, or expect idempotent behavior? Destructive operations (bootstrap-repo, teach-rule, set-enforcement) lack dry-run or confirmation patterns.
Idempotency and side-effects unclear. Tools like bootstrap-repo and teach-rule are marked WRITE but do not declare whether they are idempotent. Can an agent safely retry bootstrap-repo on a network timeout, or will it create duplicate .chameleon directories? This ambiguity makes retry logic unsafe.
Parameter descriptions lack examples and formats. E.g., 'repo' parameter accepts 'path or repo_id hash' but does not clarify the format of a repo_id hash or show a concrete example. Agents may guess incorrectly. filter_type for list-rules is 'optional filter by rule type (e.g., typescript, formatting, eslint)' but does not state if these are the only valid values or if custom types are allowed.
Tool composition risks. Read operations (list-rules, get-rules, get-idioms) should return IDs and references needed by write operations (teach-rule, set-enforcement). Not visible whether get-rules returns rule names in a format that teach-rule.rule_name can accept, or whether the agent must perform lookup/transformation.