MCP tools for AI agents
Wystan ships an MCP server that Cursor, Claude Code, Cline, and Aider call to get precise codebase context — the same data your engineering leaders see, never your raw source. Your agent stops guessing about your code.
What your agent gets
Function-level accuracy. The agent pulls the feature map and reads only the symbols that matter, instead of grepping whole directories.
Who recently contributed to the mapped area and which hotspot files the scan recorded there — without calling contributor count a bus factor.
Which mapped features and flows a diff touches, plus co-changed-but-missing files — without inventing a regression probability.
Production errors (Sentry) and usage (PostHog) joined to each feature automatically — your agent sees what's actually breaking.
Set it up
The MCP is a hosted, OAuth-secured server. Claude Code, Claude Desktop, Cursor (1.0+), VS Code (1.101+) and Codex CLI speak remote MCP natively — add the URL, sign in once in the browser, done. In Claude Code:
claude mcp add --transport http wystan https://app.wystan.ai/api/mcpFor Cursor that's ~/.cursor/mcp.json:
{
"mcpServers": {
"wystan": { "url": "https://app.wystan.ai/api/mcp" }
}
}Deployment modes (Hosted / Local / VPC) are covered on the Getting started page; runtime joins on Integrations.
The 14 tools
All read-only. They read precomputed scan fields — the engine does the work at scan time, the server just serves it. The same 14 names exist in every deployment mode.
list_featuresEvery mapped feature with its name, description, file count, flow count, and scan activity.find_featureFuzzy-match a feature by name or alias; returns its paths, flows, recorded contributors, and description.get_feature_filesThe file paths that make up a feature.get_flow_filesThe file paths that make up one flow inside a feature.get_repo_summaryStructural repository counts (features, flows, attributed files, commits) and the scan timestamp.get_hotspotsFeatures with the most engine-recorded hotspot files — a file count with the paths as receipts. Not defect risk or a fix rate.get_feature_ownersContributors recorded for a feature, with an explicit note that this is not bus factor.analyze_change_impactExact path overlap for changed files: mapped features and flows plus stored co-change receipts. No synthetic risk score.get_regression_riskCompatibility endpoint that maps touched features and abstains from returning an uncalibrated regression probability.find_symbols_in_flowPrecise functions and classes per file for a flow, with line ranges and deeplinks (falls back to file paths when symbols are absent).find_symbols_for_featureThe same symbol-level detail, aggregated across all of a feature's flows.get_feature_errorsProduction errors mapped to a feature via Sentry (hosted); returns a graceful unavailable result on local MCP.get_feature_pageviewsPostHog traffic and usage for a feature (hosted); graceful unavailable result on local MCP.get_journeys_for_fileReverse lookup from a file or line to mapped flows, journeys, owning features, and structural overlap hints.Works with your agents
Any MCP-compatible client — Cursor, Claude Code, Cline, Aider, Continue, and Claude.ai web (hosted mode). Drop the server into the config, restart the agent, and start asking questions that previously needed a senior engineer.
Always fresh
After every push an incremental scan re-maps only what changed, so the agent's context never goes stale — and PR branches get their own current scan, separate from main. The tools return precomputed, feature-grained answers, so the agent reads the symbols that matter instead of grepping whole directories.