# g-shevchenko/mcp-token-savers

Source: https://glimind.com/server/github%2Fg-shevchenko%2Fmcp-token-savers
Registry id: `github/g-shevchenko/mcp-token-savers`
Status: unknown

## Tools (80)

- `add_case_from_feedback`: Promote a reviewed retrieval/context/vision/trace miss or partial into a local benchmark case. Raw query/code is not stored; raw_query/corrected_query are hashe
- `check_doc_registry`: Compare repo docs with a repo doc registry JSON/Markdown file.
- `check_licenses`: Summarize dependency licenses from package-lock data and flag unknown/disallowed licenses.
- `compare_repo_notion_mirror`: Compare repo Markdown SSOT docs with a local Notion mirror manifest.
- `compare_runs`: Compare two local dataset run IDs and return aggregate deltas. No paths or raw case contents are returned.
- `compare_sessions`: Compare two local session rollups and return aggregate-only deltas for autonomous-loop regression review.
- `compress_context`: Query-aware or general deterministic context compression. Local-first, extractive, and API-free.
- `create_contract_snapshot`: Create a local aggregate OpenAPI/Zod/env contract snapshot for later diffing.
- `delete_first_report`: What in the diff/repo slice can be deleted without losing behavior: unreferenced new exports, dead branches, redundant guards, unused files. Tied to test presen
- `detect_ai_slop`: Deterministic AI-slop detection (CD01-CD06): abstraction bloat, guard spam, unused exports, re-export plumbing. Advisory; agents read exact files before edits.
- `diff_contracts`: Compare baseline/current contract snapshots and report removed operations, required fields, and env declarations.
- `export_dataset_manifest`: Write a local manifest artifact with case ids and counts only. It excludes expected path values and raw query text.
- `export_pantheon_safe`: Return aggregate-only agent trace telemetry for Pantheon. No raw prompts, code, paths, summaries, or artifact URLs.
- `extract_failure_step`: Extract the primary failing action, console error, or network failure from Playwright trace evidence.
- `extract_repo_actions`: Extract repo action markers without returning raw doc bodies.
- `find_broken_anchors`: Find relative Markdown anchors that do not match a target document heading.
- `find_broken_links`: Find relative Markdown links whose local file target does not exist.
- `find_change_hotspots`: Count frequently touched files over a git history window. No raw diffs or file bodies.
- `find_cochange_files`: Find files commonly changed with target paths. No raw diffs or file bodies.
- `find_duplicate_sections`: Find exact normalized duplicate doc sections without returning section bodies.
- `find_files`: Find likely relevant local repo files for a query without returning code snippets. Use as a cheap first pass when you only need file candidates.
- `find_orphan_docs`: Find docs with no inbound Markdown links from scanned docs. Advisory only; external links may exist.
- `find_references`: Find files and line numbers referencing a local symbol. Does not return source line text.
- `find_stale_notion_mirrors`: Return stale/missing Notion mirror evidence from a local mirror manifest.
- `find_symbol`: Find symbol definitions by name/query across the local graph.
- `get_artifact`: Read a local artifact produced by language-graph-mcp.
- `get_blast_radius`: Return a compact local impact set for a file or symbol based on importers and references. Use before broad edits.
- `get_file_outline`: Return symbols, imports, and importers for one indexed file. Read exact files before editing.
- `get_graph_status`: Return local graph freshness and aggregate counts. No paths or source bodies are returned.
- `get_import_neighbors`: Return imports from a file and files importing it, using best-effort local import resolution.
- `get_measurement_report`: Return local language-graph usage, quality counters, token-savings, and Pantheon-safe aggregate export.
- `get_repo_map`: Build a token-budgeted local repo map from path policy plus code-graph-lite symbols.
- `get_runtime_diagnostics`: Return metadata-only runtime diagnostics for context-prep-mcp configuration.
- `import_retrieval_feedback`: Import retrieval-mcp feedback.jsonl misses/partials into a local dataset. Raw query text is hashed only; request logs keep counts and hashes.
- `index_env_contracts`: Index .env.example-style declarations and process.env usages without reading secret values.
- `index_openapi`: Index local OpenAPI/Swagger JSON/YAML files into compact operation and schema facts.
- `index_repo`: Build or refresh a local structural language graph for a repo. Stores only symbols/imports/references metadata, never file bodies.
- `index_zod`: Index local Zod object schemas from TS/JS files using dependency-light structural parsing.
- `inventory_docs`: Inventory local Markdown/MDX docs, headings, line counts, large docs, and frontmatter gaps.
- `list_datasets`: List local golden datasets and compact counts. No case paths or raw queries are returned.
- `package_age_report`: Summarize direct dependency staleness/deprecation using optional local registry metadata.
- `prep_logs`: Compress long terminal, CI, build, test, or runtime logs into a compact debugging context.
- `prep_text`: Turn a long pasted text, meeting notes, handoff, spec, or chat history into compact carry-forward context.
- `prep_url`: Fetch and clean a public URL into compact LLM-ready markdown with title, canonical URL, headings, key facts, links, warnings, and cleaned artifact URL.
- `prepare_trace`: Parse a Playwright trace.zip, JSONL, HAR, console, or network artifact into compact browser-debug evidence and parser-stack handoff hints.
- `prepare_trace_screenshots`: Extract or copy trace screenshots into local artifacts and return vision-mcp handoff hints.
- `propose_cleanup_plan`: Combine repo hygiene scans into a reviewed cleanup plan. Advisory only; no files are changed, moved, or deleted.
- `propose_notion_update`: Create an advisory Notion update plan. Does not write to Notion.
- `record_feedback`: Record the outcome of a retrieval call so misses can become benchmark candidates.
- `record_property_run`: Append a property run result to .agent/pbt/runs.jsonl for durable audit/history. Returns generated run_id. Used for tracking regressions, comparing runs across 
- `record_step`: Record a compact metadata-only agent step in a local trace session.
- `record_tool_result`: Record a compact utility MCP/tool result in a local trace session.
- `retrieve_context`: Find the most relevant local code/docs snippets for a task before spending frontier-model tokens.
- `review_diff`: Risk-scored review of a diff: changed symbols, blast-radius size, static-check deltas, AI-slop hits. LLM reranks findings; detection is deterministic.
- `run_dataset`: Evaluate supplied result paths against a local dataset and write a local run artifact. If results are omitted, returns a needs_results run manifest.
- `run_npm_audit`: Summarize npm audit JSON. Runs npm audit only when allow_network=true; otherwise use audit_json_path/audit_json.
- `run_osv_scanner`: Summarize OSV scanner JSON. Runs osv-scanner only when allow_network=true and binary exists; otherwise use osv_json_path/osv_json.
- `run_property`: Spawn hypothesis (Python) or fast-check (TypeScript) with caller-supplied strategies + property check. Returns structured {outcome, counterexample, shrunk_input
- `run_retrieval_dataset`: Run reviewed dataset cases through local retrieval-mcp via stdio and evaluate returned ranked paths. Uses query_summary or local query_overrides; request logs k
- `scan_complexity_hotspots`: Rank local complexity hotspots by simple lines/functions/branch/import heuristics.
- `scan_dependency_cycles`: Find relative JS/TS import cycles as local advisory cleanup evidence.
- `scan_duplicate_code`: Find duplicate normalized code blocks across local source files without returning block text.
- `scan_unused_code`: Find advisory unused export candidates by local reference count. No auto-delete; agents must inspect exact files before edits.
- `scan_unused_dependencies`: Find advisory package dependency cleanup candidates from local package.json plus import/require usage. No auto-delete.
- `search_commits`: Search commit subjects/messages and return compact local history hits. No raw diffs or file bodies.
- `simplification_plan`: Ordered simplification candidates with effort/risk/behavior-preservation note. No file changes.
- `start_trace`: Start or label a local agent trace session. Stores metadata-only trace state.
- `suggest_strategies`: Maps a natural-language input description (e.g. 'positive integer', 'non-empty string', 'list of integers') to hypothesis (Python) or fast-check (TypeScript) st
- `summarize_blame`: Summarize git blame authorship for a file or line range. No source lines, raw diffs, or file bodies.
- `summarize_breaking_changes`: Convert contract diff facts into an advisory migration/review checklist.
- `summarize_console`: Summarize Playwright/browser console errors and warnings without dumping raw console logs.
- `summarize_diff_stat`: Summarize name-status and shortstat for a git range. No raw diffs or file bodies.
- `summarize_file_history`: Summarize commit history for one file with --follow. No raw diffs or file bodies.
- `summarize_lockfile_diff`: Compare npm lockfile snapshots and report added, removed, changed, and major-bump dependency facts.
- `summarize_network`: Summarize Playwright/HAR/network failures, 4xx/5xx counts, and redacted URL previews/hashes.
- `summarize_npm_audit_fix_plan`: Summarize npm audit fix dry-run JSON or stdout, including npm outputs that prepend action lines before JSON. Runs dry-run only when allow_network=true.
- `summarize_recent_commits`: Summarize recent git commits as compact metadata. No raw diffs or file bodies.
- `summarize_session`: Summarize one local agent trace session into compact counts and local artifacts.
- `summarize_supply_chain_risk`: Summarize lockfile supply-chain risk signals such as install scripts, missing integrity, git/file/link sources, and non-registry resolved packages without retur
- `validate_payload_sample`: Validate a payload sample against a JSON Schema object using local AJV. Logs only hashes/counts.

Measured by Glimind (https://glimind.com): MCP handshake + tool list only, no tool is ever called. CC BY 4.0.
