key

name	{"[object Object]":null}
description	Use the {{DISPLAY_NAME}} for code structure analysis: index repos, search symbols, trace call chains, find change blast radius, and run safe rule-based refactors. MCP-first — query the graph before reading files. Triggers on: find symbol/function/class, who calls X, impact of changing X, rename/refactor, dead code, circular deps.

{{DISPLAY_NAME}}

{{TAGLINE}}

Use this whenever you analyze code structure, locate symbols, trace dependencies, assess change impact, or run safe refactors. The server keeps a live SQLite graph of symbols and edges — always query it before reading files directly. Tools are exposed as {{TOOL_NAMESPACE}}.

Step 0 — Resolve repoId

Run list_repositories to get registered repos and their exact repoPath strings. Copy the path verbatim — never rewrite drive-letter casing or slash style (mismatches hit the path allowlist).

Step 1 — Health check

health_check(repoId: "<repoId>")

If shouldReindex: true or the repo is unknown:

index_repository(repoId, repoPath: "<exact path>", mode: "incremental", docsMode: "off", profile: "compact")

Core workflows

Find a symbol

search_symbols(repoId, query: "ExactIdentifier", strategy: "name", profile: "compact")
→ get_symbol_context_pack(repoId, name: "ExactIdentifier")   // callers + callees + change context in one call

Use strategy:"name" first; fall back to strategy:"intent" only if it returns 0. Max 2 rewrites.

Analyze change impact

find_impact_files(repoId, changedFiles: ["src/foo.ts"], depth: 2, profile: "compact")
link_tests_to_source(repoId, filePath: "src/foo.ts", minScore: 0.7)

Trace execution

search_symbols → callable symbolId
trace_execution_flow(repoId, entrySymbolId: "<id>", maxDepth: 4)
get_call_chain(repoId, symbolId: "<id>", direction: "callers", depth: 3)

Use a callable symbolId (function/method), not a class/module id.

Regex / pattern search — prefer search_regex (matches + context + enclosing symbol; scanAll:true for non-code text) over baseline grep.

Read a symbol's source — prefer get_symbol_source (by symbolId or name) over read_file; fall back to read_file only for non-symbol regions.

Safe refactor (preview → apply → rollback)

refactor_replace_preview(repoId, searchPattern, replacePattern, scope: { filePaths: ["src/**/*.ts"] })
// returns previewId, approvalToken, hunks, riskFlags — review before applying
refactor_replace_apply(previewId, approvalToken, includeLowConfidence: false)
refactor_replace_rollback(applyId)   // if needed

For renames use rename_assist(emitPreview:true) then refactor_replace_apply (includeLowConfidence:true for top-level identifiers). Approval tokens expire in 30 min — re-run preview if expired. Refactors are rule-based only (llmInvolved:false).

Guardrails

• No-LLM policy (hard): no runtime LLM calls; CODEBASE_INDEX_LLM_ENABLED=true aborts startup. Do not attempt to relax it.
• Path allowlist: only paths under CODEBASE_INDEX_ALLOWED_ROOTS are indexable. Reuse exact repoPath from list_repositories.
• Watch policy: keep watch_repo off except during active implementation/debug; stop immediately after.
• Profiles: compact (default) for most calls, nano for high-volume, standard/verbose only when you need full fields.
• Budget: soft cap 5 MCP calls/question, hard cap 8, then baseline grep/read fallback. Max 2 query rewrites.

Configuration (env)

Server entry: node {{ENTRY_PATH}}

{{ENV_TABLE}}

Tool reference

{{TOOL_LIST}}