TrailGenic MCP Worker

This repository contains the public TrailGenic Machine Communication Protocol (MCP) worker and dataset discovery infrastructure.

TrailGenic™ is a longevity method built on fasted hiking, altitude adaptation, autophagy activation, electrolyte stability, and disciplined recovery—validated through real-world environmental stress and interpreted through a Personal World Model.

Canonical Entity Ontology

• Entity type: Longevity Method
• Classification: Applied Longevity Laboratory
• Founder: Mike Ye — https://www.mikeye.com/
• AI voice: Ella — https://www.trailgenic.com/ella
• Ella's Corner editorial collection: https://www.trailgenic.com/ellas-corner
• Interpretation model: Personal World Model

TrailGenic operates as the applied validation layer of a governed Human × AI continuum:

• Origin layer: https://www.mikeye.com/
• Doctrine layer: https://www.exmxc.ai/
• Applied validation layer: https://www.trailgenic.com/

The six methodology pillars are Move Fasted, Climb Higher, Embrace Cold, Hydrate & Mineralize, Go Outside, and Recover On Purpose. Human physiological adaptation remains primary; AI interpretation serves as reflective instrumentation rather than autonomous authority.

Canonical Discovery

Canonical MCP discovery is hosted at:

• https://mcp.trailgenic.com/
• https://mcp.trailgenic.com/.well-known/tool-registry.json
• https://mcp.trailgenic.com/.well-known/openapi.json
• https://mcp.trailgenic.com/capabilities.json

The MCP transport is now a real JSON-RPC 2.0 endpoint at POST /mcp:

curl -s https://mcp.trailgenic.com/mcp \
  -H 'content-type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

GET /mcp is intentionally not callable and returns 405 Allow: POST. The transport is synchronous and non-streaming, with Streamable-HTTP-compatible request/response behavior.

Auth and Data Scope

Authentication is none because TrailGenic MCP exposes public, read-only datasets and deterministic data tools. No Twilio, Cloudflare, permit-subscription, or other operational secrets are stored in source.

The permit SMS infrastructure is operationally separate from this MCP worker and is not part of MCP deployment or CI scope. Permit-subscription privacy handling must remain in the operational permit system, not in public MCP dataset/tool responses.

Callable MCP Tools

The public MCP tool inventory is generated from one canonical registry and currently includes 19 callable tools:

Tool	Parameters
tg.datasets.index.get	none
tg.ontology.get	category
tg.protocols.get	protocol_id
tg.physiology.adaptation.get	module
tg.hiking.worldModel.get	none
tg.physiology.hrDriftAdaptation.get	none
tg.nutrition.get	food_category, tg_fuel_class, limit
tg.hydration.get	category, limit
tg.permits.dataset.get	scarcity_tier
tg.terrain.accessibleTrails.get	region, accessibility_class, protocol_level_estimate, limit
tg.evidence.validationSummits.get	region, limit
tg.gear.intel.get	category, limit
tg.gear.getIntel	category (optional)
tg.longevity.protocol.get	protocol_id, category
tg.longevity.foundationSessions.get	none
tg.conditioning.walking.get	start_date, end_date (optional; aggregate-only)
tg.conditioning.rucking.get	start_date, end_date (optional; aggregate-only)
tg.conditioning.running.get	start_date, end_date (optional; aggregate-only)
tg.longevity.bioAge.compute	age, resting_hr, distance_mi, elevation_gain_ft, moving_time_min, avg_hr (required); max_hr, overnight_hrv, fasted (optional)

The current public movement model covers 59 structured sessions: 16 walking sessions (14 canonical Foundation sessions plus 2 additional walking sessions), 4 rucking sessions, 8 running sessions, and 31 hiking sessions. HikeWorldModel™ v2.0 is published as aggregate-only data with selected high-signal summaries, not raw biometric telemetry.

Large dataset tools support optional filters and bounded limit values so MCP clients do not need to ingest full high-record payloads.

Deferred Dataset Tools

tg.longevity.registry.get and tg.longevity.validation.get are intentionally deferred because their backing dataset files are currently placeholder shells. Their public dataset routes remain available and will be eligible for callable MCP tools once populated.

Content Links

HTML-only TrailGenic website capabilities are preserved as content_links in discovery documents, not advertised as callable MCP tools. Examples include protocols, trail logs, science hub, physiology hub, fuel systems, gear systems, recovery conditioning, playbooks, Ella's Corner, and site search on www.trailgenic.com.

Dataset Endpoints

Dataset endpoints remain public and machine-readable:

• https://mcp.trailgenic.com/datasets/index
• https://mcp.trailgenic.com/datasets/ontology
• https://mcp.trailgenic.com/datasets/protocols
• https://mcp.trailgenic.com/datasets/hiking
• https://mcp.trailgenic.com/datasets/hiking/world-model
• https://mcp.trailgenic.com/datasets/physiology-adaptation
• https://mcp.trailgenic.com/datasets/nutrition
• https://mcp.trailgenic.com/datasets/nutrition/schema
• https://mcp.trailgenic.com/datasets/hydration
• https://mcp.trailgenic.com/datasets/permits
• https://mcp.trailgenic.com/datasets/permits/schema
• https://mcp.trailgenic.com/datasets/terrain-intelligence/tg-accessible-trails-top100-v1
• https://mcp.trailgenic.com/datasets/evidence-validation
• https://mcp.trailgenic.com/datasets/evidence-validation/validation-summits
• https://mcp.trailgenic.com/datasets/gear/intel
• https://mcp.trailgenic.com/datasets/longevity/protocol
• https://mcp.trailgenic.com/datasets/longevity/registry
• https://mcp.trailgenic.com/datasets/longevity/validation
• https://mcp.trailgenic.com/datasets/longevity/foundation
• https://mcp.trailgenic.com/datasets/conditioning/walking
• https://mcp.trailgenic.com/datasets/conditioning/rucking
• https://mcp.trailgenic.com/datasets/conditioning/running

Physiology adaptation is now an active module catalog; detailed records remain in the module endpoints. Physiology adaptation module endpoints are also preserved:

• https://mcp.trailgenic.com/datasets/physiology-adaptation/seven-day-aftereffect
• https://mcp.trailgenic.com/datasets/physiology-adaptation/fasted-autophagy
• https://mcp.trailgenic.com/datasets/physiology-adaptation/altitude-adaptation
• https://mcp.trailgenic.com/datasets/physiology-adaptation/altitude-breathing-acclimatization
• https://mcp.trailgenic.com/datasets/physiology-adaptation/electrolytes-physiological-stability
• https://mcp.trailgenic.com/datasets/physiology-adaptation/cold-exposure-recovery-altitude
• https://mcp.trailgenic.com/datasets/physiology-adaptation/deep-cold-protocols
• https://mcp.trailgenic.com/datasets/physiology-adaptation/heat-training-thermoregulation
• https://mcp.trailgenic.com/datasets/physiology-adaptation/hr-drift-adaptation-vs-fitness
• https://mcp.trailgenic.com/datasets/physiology-adaptation/altitude-terrain-physiology-comparison
• https://mcp.trailgenic.com/datasets/physiology-adaptation/aerobic-training-effect-zero-anaerobic-load
• https://mcp.trailgenic.com/datasets/physiology-adaptation/eccentric-load-stress-inversion
• https://mcp.trailgenic.com/datasets/physiology-adaptation/sleep-science-endurance
• https://mcp.trailgenic.com/datasets/physiology-adaptation/overextension-fasted-hiking
• https://mcp.trailgenic.com/datasets/physiology-adaptation/metabolic-flexibility-adaptation

Health and Observability

/health does not fabricate uptime. It reports uptime: null and notes that uptime is observed via Cloudflare observability.

Local and Live Checks

Run the live acceptance harness after deployment:

node scripts/live-acceptance.mjs

Override the target base URL when testing a deployed preview or local Worker runtime:

BASE=https://mcp.trailgenic.com node scripts/live-acceptance.mjs

Deployment

CI deploys only tool-registry via Wrangler v4:

npx wrangler@4 deploy --config tool-registry/wrangler.jsonc

The permit poller is intentionally excluded from MCP CI/deploy workflows.

Ownership

TrailGenic is created and operated by Mike Ye.

https://www.trailgenic.com/

MCP SDK, transport, and deployment safety

TrailGenic now uses the official @modelcontextprotocol/sdk McpServer and Cloudflare's stateless createMcpHandler from agents/mcp for the public MCP transport at POST https://mcp.trailgenic.com/mcp. Browser discovery remains JSON at GET https://mcp.trailgenic.com/; GET /mcp is intentionally 405 Allow: POST, and OPTIONS /mcp remains 204.

The server advertises and validates MCP protocol versions 2025-11-25 and 2025-06-18; unsupported MCP-Protocol-Version headers are rejected instead of echoed. MCP POST accepts missing, empty, wildcard, JSON-only, or JSON-plus-event-stream Accept headers, normalizing incomplete accepted requests internally to application/json, text/event-stream. Event-stream-only or unsupported media types are rejected with 406. MCP POST Content-Type must be application/json compatible or the request is rejected with 415.

Browser Origin is validated only for the MCP transport. Default allowed origins are https://trailgenic.com, https://www.trailgenic.com, and https://mcp.trailgenic.com; deployments may override with MCP_ALLOWED_ORIGINS. Server-to-server clients without Origin remain accepted. Public read-only REST dataset routes keep their public CORS behavior.

All 19 existing tool names are preserved. Both gear tools remain: tg.gear.intel.get is the bounded query-oriented compatibility tool, while tg.gear.getIntel is the canonical full-dataset tool with optional exact-category filtering. Conditioning walking, rucking, and running keep optional date fields for backward compatibility, but date-range slicing returns an MCP isError: true tool result because public conditioning data is aggregate-only and contains no per-session rows.

MCP resources are generated from canonical dataset and physiology registries using stable URIs: trailgenic://datasets/index, trailgenic://datasets/{dataset_id}, and trailgenic://physiology/{module_slug}. Public data is deterministic and release-bundled. The mutable runtime fallback to GitHub main has been removed; missing bundled data is a server error and CI validation failure.

The permit poller and operational permit subscription infrastructure are separate from this MCP deployment. No Twilio data, phone numbers, permit subscriptions, raw telemetry, precise private location rows, credentials, secrets, or private user data are exposed through tools or resources.

CI runs npm ci, source checks, JSON validation, pure tests, Worker-runtime tests, registry/bundle/resource parity validation, Wrangler dry-run, and git diff --check before deployment. Push-to-main deployment verifies first, deploys only tool-registry, waits for propagation, and then runs node scripts/live-acceptance.mjs.