Meta Programming¶

Meta Programming — mvp.meta_programming

Cluster: Code Intelligence | Type: component | MCP Tools: 56

Overview¶

AST-based code analysis, transformation, and generation engine supporting parse, rename, extract, and generate operations for functions, classes, and dataclasses.

When to use:

Automated code refactoring and renaming
Generating boilerplate code (functions, classes, dataclasses)
Analysing code structure and extracting function signatures

Example:

from mvp.meta_programming import MetaProgrammingBlock, CodeInput

block = MetaProgrammingBlock(name="mp")
result = block.infer(CodeInput(source="def add(a, b): return a + b", op="analyze"))
# result.ok → True; result.value → CodeOutput with functions, classes, imports

Works well with: cegis, formal_methods, adapt_healing

Production Safety Notes¶

meta_programming is suitable for local MCP pilots, trusted developer workflows, syntax checks, code summaries, scaffold generation, and assisted refactoring. It should be treated as a code-intelligence assistant, not as a hardened hosted execution boundary.

Remaining production caveats:

run_code executes Python in a subprocess with timeouts and output caps. That is useful for trusted local demos and developer workflows, but it is not a container, VM, seccomp jail, or multi-tenant sandbox. Do not expose run_code for arbitrary untrusted customer code in a hosted service without an outer isolation layer, filesystem/network restrictions, per-request quotas, and audit logging. Runtime output exposes restricted_prepass_applied; when it is false, no AST-level RestrictedPython pre-pass was applied.
Generated and refactored code is heuristic. Operations such as refactor, rename, transform_functional, generate_tests, and self_heal can produce useful drafts, but outputs must be reviewed and verified with tests before being committed or described as semantically guaranteed.
The component directory currently includes a large vendored v0.0.148 source code snapshot. Check packaging, Docker context, license notices, and install artifacts before deployment so the snapshot does not bloat MCP installs or expose unnecessary third-party code.
fetch_docs depends on Context7/MCP tooling, external network access, and monthly quota. Demo flows should handle missing SDKs, npx failures, offline environments, cache misses, and quota exhaustion with clear fallback messages.

For first-customer launch, prefer the local/trusted MCP path: analyze, generate, scan, and suggest patches; then require syntax checks, security_scan, tests, and human review before execution or merge.

Public API¶