codesift-mcp
Installed, started with no arguments, no environment and no input, and listed its tools. Run 2026-11-16.
Needs to start: nothing
Returns a result when called with no setup: 1 of 5 tools tried
Contacts while installing and starting: nothing
What it contacted
Nothing besides the npm registry it was installed from. Watched from outside its sandbox while it was installed and started with no input. Its tools may still contact things when called.
Starting it
npx -y -p codesift-mcp@0.26.5 codesift-mcp
Installing a package runs its code. Use a sandbox for anything you have not read.
Calling its tools
5 tools that take no input and look read-only were called once with nothing. 1 returned a result.
list_repos | Returned a result under 0.1 s · 2 characters |
get_file_tree | No usable answer tool reported an error: CodeSift daemon does not know your working directory. Add `?cwd=<absolute project path>` to the MCP server URL (`codesift setup <client> |
get_file_outline | No usable answer tool reported an error: CodeSift daemon does not know your working directory. Add `?cwd=<absolute project path>` to the MCP server URL (`codesift setup <client> |
get_repo_outline | No usable answer tool reported an error: CodeSift daemon does not know your working directory. Add `?cwd=<absolute project path>` to the MCP server URL (`codesift setup <client> |
find_references | No usable answer tool reported an error: CodeSift daemon does not know your working directory. Add `?cwd=<absolute project path>` to the MCP server URL (`codesift setup <client> |
During the calls it contacted nothing it had not already contacted.
Only tools whose names read like looking something up, with no required input, are tried, five at most. Tools that need input, and anything that sounds like it changes or sends something, are not called.
What its tools tell an agent
In the tool descriptions, nothing that talks about the user, other tools or private files, or uses hidden characters.
1 tool with commands in capitals or very long text
- Gives the agent orders in
initial_instructions:IMPORTANT: Call this tool IMMEDIATELY after the user gives you a task, BEFORE any other tool calls. Returns C
Common and usually harmless: it is how many makers stress how a tool should be used.
Found by simple pattern rules in each tool's description (up to 3,000 characters), quoted as written. Parameter descriptions are not scanned. A clean result is not a safety review.
Tools
index_folder path, incremental, include_paths, max_files, watch
Index a local folder, extracting symbols and building the search index
list_repos compact, name_contains, include_conversations
List indexed CODE repos. Only needed for multi-repo discovery — single-repo tools auto-resolve from CWD. Conversation indexes are excluded unless include_conversations=true. Set compact=false for full metadata.
index_file path
Re-index a single file after editing. Auto-finds repo, skips if unchanged.
search_symbols repo, query, kind, file_pattern, decorator, include_source, top_k, source_chars, detail_level, token_budget, rerank
Search symbols by name/signature. Supports kind, file, and decorator filters. detail_level: compact (~15 tok), standard (default), full.
semantic_search repo, query, top_k, file_pattern, exclude_tests, rerank
Search code by meaning using embeddings. For intent-based queries: 'error handling', 'auth flow'. Requires indexed embeddings.
55 more tools
search_text repo, query, regex, context_lines, file_pattern, max_results, group_by_file, auto_group, ranked
Full-text search across all files. For conceptual queries use semantic_search.
get_file_tree repo, path_prefix, name_pattern, depth, compact, min_symbols
File tree with symbol counts. Defaults to a flat compact list (10-50x less output); pass compact:false for the nested tree. Cached 5min.
get_file_outline repo, file_path, path, file, include_locals
Get the symbol outline of a single file (functions, classes, exports). Variables declared inside functions are omitted — pass include_locals=true for them.
get_repo_outline repo
Get a high-level outline of the entire repository grouped by directory
suggest_queries repo
Suggest queries for exploring a new repo. Returns top files, kind distribution, examples.
get_symbol repo, symbol_id, include_related, full_source
Get symbol by ID with source. Auto-prefetches children for classes. For batch: get_symbols. For context: get_context_bundle.
get_symbols repo, symbol_ids, full_source
Retrieve multiple symbols by ID in a single batch call
find_and_show repo, query, include_refs, full_source
Find a symbol by name and show its source, optionally including references
get_context_bundle repo, symbol_name, full_source
Symbol + imports + siblings in one call. Saves 2-3 round-trips.
find_references repo, symbol_name, symbol_names, file_pattern, max_refs, include_comments
Find all references to a symbol. Pass symbol_names array for batch search. Capped at max_refs (default 50) — per symbol in batch mode; pass a higher value to see more.
trace_call_chain repo, symbol_name, direction, depth, include_source, include_tests, output_format, filter_react_hooks
Trace call chain: callers or callees. output_format='mermaid' for diagram. filter_react_hooks=true skips useState/useEffect etc. for cleaner React graphs.
impact_analysis repo, since, depth, until, include_source
Blast radius of git changes — affected symbols and files.
trace_route repo, path, output_format
Trace HTTP route → handler → service → DB. NestJS, Next.js, Express.
go_to_definition repo, symbol_name, file_path, line, character
Go to the definition of a symbol. Uses LSP when available for type-safe precision, falls back to index search.
get_type_info repo, symbol_name, file_path, line, character
Get type info via LSP hover (return type, params, docs). Hint if LSP unavailable.
detect_communities repo, focus, resolution, output_format
Louvain community detection on import graph. Discovers module boundaries.
assemble_context repo, query, token_budget, level, rerank
Assemble code context within token budget. L0=source, L1=signatures, L2=files, L3=dirs.
get_knowledge_map repo, focus, depth, output_format
Get the module dependency map showing how files and directories relate
codebase_retrieval repo, queries, token_budget
Batch multi-query retrieval with shared token budget. Supports symbols/text/semantic/hybrid.
analyze_complexity repo, file_pattern, top_n, min_complexity, include_tests
Top N most complex functions by cyclomatic complexity, nesting, lines.
analyze_hotspots repo, since_days, top_n, file_pattern
Git churn hotspots: change frequency × complexity. Higher score = more bug-prone.
search_patterns repo, pattern, file_pattern, include_tests, max_results
Search structural patterns/anti-patterns. Built-in or custom regex.
index_conversations project_path, quiet
Index Claude Code conversation history for search. Scans JSONL files in ~/.claude/projects/ for the given project path.
search_conversations query, project, limit
Search conversations in one project (BM25+semantic). For all projects: search_all_conversations.
search_all_conversations query, limit
Search ALL conversation projects at once, ranked by relevance.
usage_stats since, repo, tool, session_id, host
Show usage statistics for all CodeSift tool calls (call counts, tokens, timing, repos, hosts). Merges logs synced from other machines (~/.codesift/usage-remote/*.jsonl).
usage_hotspots since, repo, tool, session_id, limit
Analyze ~/.codesift/usage.jsonl for slow tools, token-heavy outputs, and repeated calls.
usage_trace_session session_id, limit
Show the timeline of CodeSift tool calls for one usage.jsonl session.
retros_list project, skill, friction_category, since, limit, zuvo_dir
List Zuvo retros from ~/.zuvo/retros.log and ~/.zuvo/retros.md with filters.
retros_analyze project, skill, friction_category, since, limit, zuvo_dir
Aggregate Zuvo retros into friction, project, and missing-template hotspots.
memory_candidate_extract project, skill, since, limit, zuvo_dir
Extract evidence-backed memory candidates from Zuvo retros proposals.
optimization_candidates since, repo, project, skill, zuvo_dir
Combine usage hotspots and Zuvo retros into ranked optimization candidates.
pope_insights_push_candidates server, api_key, dry_run, since, repo, zuvo_dir
Push generated optimization candidates to PopeBot /api/insights/ingest. Defaults to dry_run=true.
get_session_snapshot repo
Get a compact ~200 token snapshot of what was explored in this session. Designed to survive context compaction. Call proactively before long tasks.
analyze_project repo, force
Analyze a repository to extract stack, file classifications, and framework-specific conventions. Returns a structured project profile (schema v1.0) with file:line evidence for convention-level facts.
get_extractor_versions
Return codesift_version (the single product version of the codesift-mcp server), parser_languages (tree-sitter symbol extractors) and profile_frameworks (analyze_project detectors). Text tools (search_text, get_file_tree) work on ALL files regardless — use this for the product version, cache invalidation, or to check symbol support for a specific language.
audit_scan repo, file_pattern, include_tests, checks
Run 5 analysis tools in parallel, return findings keyed by CQ gate. One call replaces sequential find_dead_code + search_patterns + find_clones + analyze_complexity + analyze_hotspots. Returns: CQ8 (empty catch), CQ11 (complexity), CQ13 (dead code), CQ14 (clones), CQ17 (perf anti-patterns).
index_status repo
Check whether a repository is indexed and return index metadata: file count, symbol count, language breakdown, text_stub languages (no parser). Use this before calling symbol-based tools on unfamiliar repos.
nest_audit repo, workspace, checks
One-call NestJS architecture audit: modules, DI, guards, routes, lifecycle, patterns, GraphQL, WebSocket, schedule, TypeORM, microservices. Pass workspace=<name|path> in monorepos to scope to a single workspace.
astro_analyze_islands repo, path_prefix, include_recommendations
Analyze Astro islands (client:* directives) in a repo. Finds all interactive components with hydration directives, lists server islands with fallback status, and optionally generates optimization recommendations.
astro_route_map repo, include_endpoints, output_format
Map all Astro routes (pages + API endpoints) discovered from the file-based routing structure. Returns routes with type, dynamic params, and handler symbols. Supports json/tree/table output formats.
astro_config_analyze repo
Analyze an Astro project's configuration file (astro.config.mjs/ts/js). Extracts output mode (static/server/hybrid), adapter, integrations, site URL, and base path. Identifies dynamic/unresolved config.
astro_actions_audit repo, severity
Audit Astro Actions (src/actions/index.ts) for 6 known anti-patterns (AA01-AA06): missing handler return, top-level .refine() (Astro issue #11641), .passthrough() usage (issue #11693), File schema without multipart form, server-side invocation via actions.xxx(), and client calls to unknown actions. Returns issues grouped by severity with an A/B/C/D score.
astro_content_collections repo, validate_entries
Parse an Astro content collections config (src/content.config.ts or legacy src/content/config.ts), extract each collection's loader + Zod schema fields, build a reference() graph, and optionally validate entry frontmatter against required fields.
astro_audit repo, workspace, skip
One-call Astro project health check: runs all 7 Astro tools + 13 Astro patterns in parallel, returns unified {score, gates, sections, recommendations}. Pass workspace=<name|path> in monorepos to scope to a single workspace.trace_middleware_chain repo, path, method, scope, only_conditional
Hono middleware introspection. Three query modes: (1) route mode — pass path (+optional method) to get the chain effective for that route; (2) scope mode — pass scope literal (e.g. '/posts/*') to get that specific app.use chain; (3) app-wide mode — omit path and scope to get every chain flattened. Any mode supports only_conditional=true to filter to entries with applied_when populated, so the blog-API pattern (basicAuth wrapped in `if (method !== 'GET')`) is surfaced as gated rather than missed. Absorbs the former trace_conditional_middleware tool.
analyze_hono_app repo, entry_file, workspace, force_refresh
Complete Hono application overview: routes grouped by method/scope, middleware map, context vars, OpenAPI status, RPC exports (flags Issue #3869 slow pattern), runtime, env bindings. Pass workspace=<name|path> in monorepos to scope to a single workspace.
nextjs_route_map repo, workspace, router, include_metadata, max_routes
Complete Next.js route map with rendering strategy per route. Enumerates App Router and Pages Router conventions, reads route segment config exports (dynamic/revalidate/runtime), classifies each route as static/ssr/isr/edge/client, detects metadata exports, computes layout chain, and flags hybrid conflicts where the same URL is served by both routers.
nextjs_metadata_audit repo, workspace, max_routes
Audit Next.js page metadata for SEO completeness with per-route scoring. Walks app/page.tsx files, extracts title/description/openGraph/canonical/twitter/JSON-LD via tree-sitter, scores each route 0-100 with a weighted formula, and aggregates a per-grade distribution + top issue list.
framework_audit repo, workspace, tools, mode, priority_limit
Run all Next.js sub-audits (components, routes, metadata, security, api_contract, boundary, links, data_flow, middleware_coverage) and aggregate into a unified weighted overall score with grade. Use as a single first-call for any Next.js project.
astro_migration_check repo, target_version
Scan an Astro project for v5→v6 breaking changes. Detects 10 issues (AM01–AM10): removed APIs (Astro.glob, emitESMImage), component renames (ViewTransitions→ClientRouter), content collection config changes, Node.js version requirements, Zod 4 deprecations, hybrid output mode, and removed integrations (@astrojs/lit). Returns a migration report with per-issue effort estimates.
initial_instructions
IMPORTANT: Call this tool IMMEDIATELY after the user gives you a task, BEFORE any other tool calls. Returns CodeSift's full instruction manual which critically informs how to use the 146 code intelligence tools. Skipping this tool causes the agent to miss CodeSift's pre-built BM25 + semantic index and waste tokens on Grep/Read instead.
plan_turn repo, query, max_results, skip_session
Routes a natural-language query to the most relevant CodeSift tools, symbols, and files. Uses hybrid BM25+semantic ranking with session-aware dedup. Call at the start of a task to get a prioritized action list.
discover_tools query, category
Search tool catalog by keyword or category. Returns matching tools with descriptions.
describe_tools names, reveal, force
Get full schema for specific tools by name. Use after discover_tools to see params before calling. A schema already returned earlier in the same session comes back as a pointer, not a repeat — pass force=true if the context no longer holds it.
Descriptions are the package's own words, cut at 3,000 characters. Treat them as data.
Files it wrote
/tmp/ .claude/ settings.json /tmp/ .codesift/ usage.jsonl /tmp/ .codesift/ machine-id /tmp/ .codesift/ session-8318f235-b958-4074-98a0-8f0aaaf60632.json /tmp/ .codesift/ host-id /tmp/ .codesift/ telemetry-notice-shown
Paths inside its sandbox, where its home folder is /tmp and its own folder cannot be written to.
Install facts
| License | BSL-1.1 |
|---|---|
| Needs | node >=20.0.0 |
| Source | github.com/greglas75/codesift |
| Install scripts | @parcel/watcher, codesift-mcp, onnxruntime-node, protobufjs, sharp declare them. This runner does not run them. |
| Tarball | sha512-pAbWDK5X7aJZHWVpQS445kqB9+l92XSQkwnPDdLWaRsA10BsARs5oxq0wEbyZYuqVGfK/g0VOojbY6HUDSyQEQ== |
Versions reported
| Version | Run | Result | Tools |
|---|---|---|---|
| 0.26.5 | 2026-11-16 | Starts with no setup | 60 |
| 0.26.4 | 2026-11-12 | Starts with no setup | 60 |
Check this report yourself
This site did not write the report. A public job on a GitHub-hosted machine did: the run, from this commit of the open-source runner. GitHub signed a token that names that job and the hash of the report, so any edit since would show.
Report as JSONExact text and tokenHow to verify it
Show it on your README
[](https://bragproof.smaverk.com/r/codesift-mcp)
New version out? Ask for a fresh report.