MCP Tools
Everything your AI needs to understand your codebase.
Detailed documentation for each MCP tool
Semantic code search by meaning
Semantic code search that finds code by meaning rather than exact keyword matches. Instead of searching for "auth_handler", you can ask "where do we handle JWT validation?" and get the relevant files—not just string matches.
Know what you want but not the exact filename or function name? Semantic search understands code semantics, not just text. Perfect for exploring unfamiliar codebases or finding implementation patterns.
Index your codebase for semantic search
Builds a searchable vector index of your entire codebase. Once indexed, semantic queries return results in ~50ms instead of scanning every file on each search.
This is the one-time setup that enables fast semantic search. Without indexing, every search would scan your entire codebase. With it, searches are instant.
View indexing statistics and health
Displays indexing statistics and health metrics: file count, chunk sizes, last update time, index size, and any errors during indexing.
Transparency into the indexing process. Helps diagnose why certain files might not appear in search results or if there are issues with your index.
Track and optimize token usage
Tracks token usage across queries. Shows how many tokens are being used per query, how much context is being injected, and calculates potential savings from context optimization.
Understand the ROI of context optimization. See exactly how many tokens you're saving with semantic search versus naive approaches.
Resume sessions with full context
Resumes sessions with full context memory—brings back what was discussed, what files were examined, and what was learned in previous conversation turns.
AI tools typically lose context between sessions. This preserves the knowledge from your work session so you don't have to re-explain your codebase every time.
Generate session activity summaries
Generates summaries of session activity: what files were read, what was discussed, what was modified, and key decisions made during the session.
Get a quick overview of what happened in a work session. Perfect for onboarding to resume work, or sharing context with teammates.
Track file edits for context coherence
Tracks file edits for context coherence. When you edit a file, this tool updates the system's awareness so search results remain accurate.
After making code changes, your index may become stale. Registering edits keeps the context aware of changes without manual re-indexing.
Explore code dependencies and connections
Explores code dependencies and connections—shows what imports what, what calls what, and the relationships between files and modules in your codebase.
Understand the relationships in your codebase at a glance. When editing a core module, see everything that depends on it before making changes.
Analyze blast radius of code changes
Analyzes the "blast radius" of proposed code changes—shows how many files might be affected downstream if you make a particular change.
Before refactoring, understand the scope of impact. Make informed decisions about whether a change is safe, or what else needs updating.
Index an Obsidian vault for semantic retrieval
Indexes all .md notes in an Obsidian vault using PyckLM embeddings. The indexed notes become searchable via search_code, letting any MCP client query your personal knowledge base alongside code.
Find design decisions, meeting notes, and architecture docs by meaning — without leaving your editor. Your Obsidian vault becomes part of your code context.
Index a Notion database or page for semantic queries
Exports a Notion database or page to markdown via the Notion API, then indexes the content with PyckLM. Supports single pages, full databases (up to 500 pages), and nested content up to 5 levels deep.
Query your Notion project specs, onboarding docs, and product requirements by natural language directly from your AI editor. Keep product context in sync with code.
Query commit history with natural language
Indexes git commit messages, diffs, and co-change patterns from any local repository. Commits are stored alongside code chunks and fully searchable via search_code.
Find when a bug was introduced, discover which files change together, or recall what work touched a specific area — without git blame archaeology.
Model-agnostic prompt routing for any MCP client
The editor-side equivalent of Pyckle's Claude Code hook. Given a user prompt, it decides whether code context is needed and injects the most relevant chunks — automatically. Works in Cursor, Windsurf, Continue, VS Code, and Zed.
Pyckle's automatic context injection without needing Claude Code or a hook. Any MCP-compatible editor gets the same intelligence as Claude Code users.
Start an autonomous goal-directed iteration session
Initializes a structured improvement loop with a measurable goal. Stores the loop state (goal, metric command, scope, baseline) locally so Pyckle can track progress across many iterations.
Autonomous optimization of any metric — test coverage, benchmark score, bundle size, error rate — with a full audit trail of what was tried and what worked.
~/.pyckle-mcp/loops/ for durability across sessions
higher_is_better and lower_is_better directions
Record an iteration result — keep, discard, or crash
Appends an iteration entry to the active loop's JSONL log. Tracks the metric value, commit hash, status, and description for every attempt, updating the running best automatically.
Maintain a structured record of which changes improved the metric and which didn't — essential for autonomous iteration loops and post-analysis audits.
current_best on keep status
Get live progress and metrics for an active loop
Returns a markdown summary of a loop's current state: baseline → best improvement, keep/discard/crash counts, and a table of recent iterations with metrics and statuses.
Check how an autonomous session is performing mid-run, or audit results before deciding to continue, pause, or stop the loop.
List all autoloop sessions for a codebase
Lists every autoloop session run against a given codebase, sorted by most recent first. Shows goal, status, baseline-to-best improvement, and iteration counts for each loop.
Review past autonomous sessions, compare outcomes across different optimization goals, or resume a previous loop from where it left off.
Free tier includes core search and indexing. Upgrade to Pro for session memory, dependency graphs, autoloop, and more.