Try Pyrox MCP

Pyrox exposes a public, read-only MCP server at:

https://pyrox-api.fly.dev/mcp/

Use it when you want Claude, Codex, or another MCP client to answer natural-language questions about HYROX Race Results without downloading the DuckDB artifact or writing Python. The connector computes analytics server-side; it does not expose raw SQL, arbitrary row exports, or write operations.

Add It To Claude Code

Register the connector:

claude mcp add --transport http pyrox https://pyrox-api.fly.dev/mcp/

Verify it is available:

claude mcp list

By default this registers the connector for the current project. To make it available across your Claude Code projects:

claude mcp add --transport http --scope user pyrox https://pyrox-api.fly.dev/mcp/

Remove it later with:

claude mcp remove pyrox

Add It To Claude (claude.ai and Desktop)

You can also use Pyrox as a custom connector in the Claude web app and Claude Desktop, not just the CLI.

Custom connectors require a paid Claude plan (Pro, Max, Team, or Enterprise). On Team/Enterprise an admin may need to enable custom connectors first.

  1. Sign in at claude.ai (or open Claude Desktop).
  2. Go to Settings -> Connectors.
  3. Scroll down and click Add custom connector.
  4. Set:
  5. Name: Pyrox
  6. URL: https://pyrox-api.fly.dev/mcp/ (keep the trailing slash)
  7. Click Add. The server is open and read-only, so there is no sign-in or OAuth step — it connects straight away and discovers the Pyrox tools.
  8. In a chat, enable Pyrox from the tools menu in the composer, then prompt it by name, e.g. "Using Pyrox, what seasons and divisions are available?". The first tool call asks you to approve it.

Add It To Codex

Codex can use the same streamable-HTTP MCP endpoint. Register the connector:

codex mcp add pyrox --url https://pyrox-api.fly.dev/mcp/

In the Codex TUI, use /mcp to check that the server is available.

You can also configure it directly in ~/.codex/config.toml:

[mcp_servers.pyrox]
url = "https://pyrox-api.fly.dev/mcp/"

Codex shares MCP configuration between the CLI and IDE extension, so the same setup works across both.

What It Can Answer

The MCP server exposes ten tools:

Tool Use it for
list_filters Discover available seasons, divisions, genders, locations, and age groups.
list_races Find valid season and location pairs before asking for race-level stats.
find_athlete Resolve an athlete name to candidate Results with result_id values.
get_race_report Inspect one Result's full Race report and Segment breakdown.
get_athlete_profile Summarize an athlete's historical Results and personal bests.
get_distribution Get a Cohort Distribution as histogram bins and summary stats.
get_race_summary Get timing stats for all Segments in one Race.
get_cohort_segment_averages Compare Run and Station averages for all, top-N, or bottom-N athletes in one Race.
get_rankings Query leaderboard rows and hypothetical target-time placement.
get_deepdive Compare one Result against cross-location Cohorts for a metric.

Example Prompts

Start by discovering the available filters:

Using Pyrox, what seasons, divisions, genders, and locations are available?

Find an athlete and inspect one Result:

Using Pyrox, find results for Vlad Matei. Pick the most recent open Result and
summarize the Race report, including strongest and weakest Segments.

Compare a target finish time:

Using Pyrox, where would a 62-minute male open time rank in season 8?

Ask for a Cohort Distribution:

Using Pyrox, show the Distribution of female open finish times in season 8.
State the Cohort, sample size, and whether the sample is thin.

Build a profile:

Using Pyrox, create an athlete profile for Your Name and explain how his
Segment strengths compare with his historical Results.

How To Read The Answers

Pyrox uses these terms consistently:

  • Race: one HYROX event instance, identified by season, location, and year.
  • Result: one athlete's performance in one Race.
  • Segment: one timed Run or Station.
  • Roxzone: transition time between Segments.
  • Cohort: the reference population used for rankings, Distributions, and Percentiles.
  • Distribution: histogram bins plus summary stats for a metric across a Cohort.

The result_id is an internal identifier. Users do not need to know it in advance. For common or ambiguous names, find_athlete returns candidate races so Claude can choose the right Result or ask for a narrower filter. i.e Vlad Matei and Vlad Mateei (typo when registering)

Caveats

  • Pyrox is based on public HYROX source data and may inherit source-data issues.
  • Thin Cohorts are flagged with sample-size information and should be interpreted cautiously.
  • Divisions are not pooled by default because open, pro, doubles, and relay Results are not directly comparable.
  • Distributions default to the latest season and the open division when those filters are omitted.
  • Claude renders any charts from returned structured data; the server does not return image files.
  • The public MCP endpoint is open and read-only. It is rate-limited to protect availability.