Modjo Pull

/modjo-pull

Wraps the Modjo API. Modjo doesn't have a native MCP server; this skill calls the API directly and saves results in a consistent shape.

Inputs

• date_range: ISO 8601 date range (e.g. 2026-04-01:2026-04-15)
• topic: free-text topic filter passed to the Modjo search endpoint (e.g. "quote tracking")

Authentication

Reads MODJO_API_TOKEN from the local .env file. This token is not committed to the repo — set it locally before running.

Procedure

1. POST to https://api.modjo.ai/v1/calls/search with { from, to, query }.
2. For each match, fetch /calls/{id} and /calls/{id}/transcript.
3. Save each call as data/modjo/{date}/{call-id}.json with:
   • summary — Modjo's AI summary
   • transcript — full transcript with speaker turns and timestamps
   • participants — array of { name, role, company }
   • date — ISO datetime
   • language — nl | en | mixed
4. Write a data/modjo/{date}/manifest.json listing what was pulled.

Notes

• Trust the transcript, not the summary for verbatim quotes. Modjo's summarization paraphrases — sometimes substantively. The summary is a navigational aid; the transcript is the source of truth.
• Mixed Dutch/English is normal. Don't translate at pull time — translation happens at extraction so we keep the original alongside the translation.
• In workshop contexts the data is pre-pulled into data/modjo/2026-04-15/. Only re-run this skill if you need fresher calls.

API-over-MCP pattern

This is the pattern to reach for when a tool doesn't have a native MCP. Wrap the API as a skill, save results to a consistent path under data/, and treat the skill as if it were an MCP. From the user's perspective the experience is identical — /modjo-pull just works. Other candidates for the same treatment: tools where MCPs are still maturing, or where the team needs auth scoped to the repo rather than to a desktop session.