Skip to content

Get Context Bundle

Jump to bottom
jgravelle edited this page Jun 19, 2026 · 3 revisions

Get Context Bundle

Get a symbol's definition and its file's imports in one call — nothing more.

What

When an AI needs to modify a function, it typically has to load the entire file to understand its dependencies. For large files this wastes thousands of tokens on code the AI will never touch. get_context_bundle returns exactly two things: the symbol's full source and every import/require statement from the same file. That's the minimum viable context for understanding and editing any symbol.

The tool is intentionally scoped to definition + imports for v1. No call graph, no transitive dependencies — just the scaffolding a symbol needs to make sense.

Why It Matters

For everyone

If you ask an AI to fix a bug in one function, it no longer has to read the whole 800-line file first. You get faster, cheaper answers with less noise.

For engineers

Dimension Detail
Tool name get_context_bundle
Inputs repo (string), symbol_id (string)
Outputs symbol_id, name, kind, file, line, end_line, signature, docstring, source, imports[], _meta
Import extraction Language-aware regex; 16 languages supported
Go block imports import (...) blocks captured as a unit, including the open/close parens
Token savings Reported in _meta.tokens_saved vs full-file baseline
Error handling Returns {"error": "..."} on bad repo or unknown symbol ID

Supported languages for import extraction: Python, JavaScript, TypeScript, TSX, Go, Rust, Java, Kotlin, C#, C, C++, Swift, Ruby, PHP, Elixir, Scala, Haskell, Lua, Dart.

How

Usage

{
  "tool": "get_context_bundle",
  "arguments": {
    "repo": "local/my-project",
    "symbol_id": "src/auth/token.py::verify_token#function"
  }
}

What comes back

{
  "symbol_id": "src/auth/token.py::verify_token#function",
  "name": "verify_token",
  "kind": "function",
  "file": "src/auth/token.py",
  "line": 24,
  "end_line": 41,
  "signature": "def verify_token(token: str, secret: str) -> dict",
  "docstring": "Verify a JWT and return its payload.",
  "source": "def verify_token(token: str, secret: str) -> dict:\n    ...",
  "imports": [
    "import jwt",
    "import time",
    "from typing import Optional",
    "from .exceptions import TokenExpiredError"
  ],
  "_meta": {
    "timing_ms": 4.1,
    "tokens_saved": 812,
    "total_tokens_saved": 7041823,
    "cost_avoided": {
      "claude_sonnet_4_6": 0.0024
    }
  }
}

Where the change lives

  • src/jcodemunch_mcp/tools/get_context_bundle.py — new tool: resolves repo, fetches symbol source via byte-offset read, extracts imports from cached file content using language-aware regex patterns
  • src/jcodemunch_mcp/server.py — import added; tool registered in list_tools(); handler added in call_tool()
  • tests/test_get_context_bundle.py — 34 tests: unit tests for _extract_imports across 9+ language classes, integration tests via index_folder + real symbol lookup, error case coverage

Connection to the jMRI Spec

get_context_bundle is the first implementation of the context bundle retrieval pattern described in the jMRI spec — a scoped, symbol-anchored bundle that returns only what's needed to act on a symbol. Future iterations may expand the bundle (e.g. direct callers, type definitions) while keeping the same interface.

jcodemunch-mcp on GitHub · 900+ stars

Clone this wiki locally