Getting Started
Two install paths. Same server. Choose what fits your workflow.
Quick Start
npx @copilotkit/pathfinder init
npx @copilotkit/pathfinder serveQuickest start. No Docker needed. Add Postgres later for semantic search.
docker pull ghcr.io/copilotkit/pathfinder
docker run -v ./pathfinder.yaml:/app/pathfinder.yaml \
-v ./docs:/app/docs -p 3001:3001 \
ghcr.io/copilotkit/pathfinderProduction-ready. Use docker-compose to add Postgres for full semantic search.
Try It — Pathfinder on Its Own Docs
This documentation site is indexed by a live Pathfinder instance. Connect your AI agent to search and explore these docs via MCP:
Claude Code
Claude Desktop / Cursor / Any MCP Client
Once connected, your agent can search-docs("how to configure sources") or explore-docs("cat /docs/quickstart.mdx") to get answers from these pages.
Auto-Generated Endpoints
Pathfinder automatically generates and serves several discovery endpoints. These are cached and invalidated whenever sources are reindexed.
/llms.txt— A concise index of all indexed documentation, following the llms.txt convention. Useful for agents that want a quick overview of available content./llms-full.txt— The complete content of all indexed docs in plain text. Larger, but gives agents full context without multiple tool calls./.well-known/skills/default/skill.md— A dynamic agent capability description that reflects your current tool configuration. Agents can read this to understand what Pathfinder can do./faq.txt— Q&A pairs from FAQ-category sources (Slack threads, Discord forums), filtered by confidence. Only generated when sources withcategory: faqare configured./health— JSON health check with uptime, indexing status, and chunk counts per source./analytics— Built-in analytics dashboard with query logging, top queries, empty result tracking, and latency metrics. Requiresanalytics.enabled: truein config.
Analytics API Endpoints
When analytics is enabled, the following API endpoints are available (authenticated via ANALYTICS_TOKEN bearer token):
/api/analytics/summary— Overview of query volume, average latency, and top queries./api/analytics/queries— Paginated list of all logged search queries with timestamps, latency, and result counts./api/analytics/empty-queries— Queries that returned zero results, useful for identifying content gaps.
All HTTP responses include Link headers pointing to /llms.txt and, when available, </faq.txt>; rel="faq", so clients can discover these endpoints automatically.
CLI Commands
pathfinder init— Scaffold a newpathfinder.yamlconfig file interactively.pathfinder init --from <url>— Crawl a documentation site and auto-generate apathfinder.yamlconfig. Options:--rate-limit <ms>— Milliseconds between requests (default: 500)--max-pages <n>— Maximum pages to crawl (default: 500)--force— Overwrite existingpathfinder.yaml
pathfinder serve— Start the MCP server. Uses PGlite if noDATABASE_URLis set. Options:-p, --port <number>— Override the default port (3001)-c, --config <path>— Path to pathfinder.yaml config file
pathfinder validate— Validate config schema, check that required environment variables are set, and probe source connectivity. Returns pass/fail per source with actionable error messages. Exits with code 1 on errors.
Learn More
Configuration Reference
Full annotated pathfinder.yaml — server, sources, tools, embedding, indexing, and webhooks. Includes example configs for bash-only, full-stack, multi-repo, and multi-version setups.
Deployment Guide
Docker, Docker Compose, Railway, and generic VPS deployment. Environment variables, volume mounts, and production setup.
MCP Client Setup
Connect Claude Desktop, Cursor, Codex, VS Code (Continue), or any Streamable HTTP client to your Pathfinder instance.
Search Features
How semantic search works, the qmd command, related files, grep-miss suggestions, and search tool parameters.