fastcontext-explorer

FastContext Explorer

Workflow

1. Check availability first with the MCP tool fastcontext_health.
2. If health is usable, call fastcontext_explore with a specific query and the repository root.
3. Treat returned citations as candidate evidence, not proof. Read the cited files and line ranges yourself before editing or answering.
4. If citations are sparse, off-target, or missing, refine the query once with concrete terms from the task, error, subsystem, or nearby filenames.
5. For debugging poor searches, use fastcontext_explore_with_trace and inspect the saved trajectory.

When to Use

• Understand unfamiliar code before editing, reviewing, debugging, or explaining it.
• Trace logic across layers, such as request to handler to service to storage.
• Answer code questions like "How does X work?", "Where is Y defined?", or "What calls Z?".
• Map what a symbol depends on or what depends on it.
• Assess which files are likely affected by a change.

When unsure and the answer needs more than one file, run FastContext first.

When Not to Use

• You already read the exact file and surrounding context in this session.
• One known file and one obvious search will answer the question.
• The task is pure writing or generation with no repository exploration.
• The task is non-code.

Query Shape

Write concise exploration requests that name behavior and intent:

Find where OAuth access tokens are refreshed and which tests cover expired token handling.

Prefer subsystem and failure language over generic searches:

Locate the request validation path that rejects oversized uploaded files.

Avoid asking FastContext to solve or patch:

Fix the upload bug.

MCP Tools

Use fastcontext_explore:

{
  "repo_path": "/absolute/path/to/repo",
  "query": "Locate the request validation path that rejects oversized uploaded files.",
  "max_turns": 6,
  "citation": true
}

Use fastcontext_explore_with_trace when a trajectory is needed: