Connect the Pyrox MCP server¶

The public endpoint is:

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

It runs typed reporting tools over the HYROX dataset. Clients can't send raw SQL, export arbitrary rows or write data. The service does the calculation and returns structured fields for the assistant to explain or chart.

Claude Code¶

Add Pyrox to the current project:

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

Check the connection:

claude mcp list

Use --scope user if you want the connection in every Claude Code project:

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

Remove it with claude mcp remove pyrox.

Claude web and Desktop¶

Custom connectors require a paid Claude plan. A Team or Enterprise admin may need to allow them first.

Open Settings, then Connectors.
Select Add custom connector.
Enter Pyrox as the name.
Paste https://pyrox-api.fly.dev/mcp/, including the trailing slash.
Add the connector. Pyrox doesn't use OAuth or ask for an API key.
Enable Pyrox from the chat composer before sending a question. Claude asks you to approve the first tool call.

Codex¶

Register the same HTTP endpoint:

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

Run /mcp in the Codex TUI to inspect the connection. You can also add it to ~/.codex/config.toml yourself:

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

The CLI and IDE extension read the same Codex configuration.

MCP tools¶

Tool	Returns
`list_filters`	Available seasons, divisions, genders, locations and age groups.
`list_races`	Valid season and location pairs.
`find_athlete`	Candidate results for an athlete name, including `result_id`.
`get_race_report`	One result with every available run and station split.
`get_athlete_profile`	Historical results and personal-best fields.
`get_distribution`	Histogram bins and summary statistics for a cohort.
`get_race_summary`	Segment timing statistics for one race.
`get_cohort_segment_averages`	Run and station averages for a selected group of finishers.
`get_rankings`	Leaderboard rows or the placement of a hypothetical finish time.
`get_deepdive`	One result compared with cross-location cohorts for a chosen metric.

Questions to try¶

Start with the available data when you don't know the exact filters:

Using Pyrox, which season 8 locations have male open results?

Find an athlete before asking for a race report:

Using Pyrox, find results for Vlad Matei. Use the most recent open result and
report the two station splits furthest from his cohort median.

A target time doesn't need an athlete record:

Using Pyrox, where would a 62-minute male open time rank in season 8? State the
cohort size and filters with the answer.

For a distribution, name the metric and cohort:

Using Pyrox, return the distribution of female open finish times in season 8.
Include the sample size and histogram bins.

Result IDs and ambiguous names¶

A result_id identifies one performance in one race. You don't need to know it beforehand: find_athlete returns candidate records, and the assistant can pick one or ask you to narrow the name, season or location.

Names come from registration data, including misspellings. Search results may therefore contain several spellings for the same person; check the race history before treating them as one athlete.

Service limits¶

The source data may contain wrong names, divisions or split times.
Small cohorts carry sample-size warnings and need cautious interpretation.
Omitted distribution filters default to the latest season and open division.
The endpoint allows 60 requests per minute per IP and may need a few minutes to start after sitting idle.
Chart images come from the MCP client. Pyrox returns structured data rather than image files.