---
title: BEMA agent instructions
description: When and how AI agents should use BEMA.
canonical: https://bema.jpamorgan.com/agents.md
last-updated: 2026-08-22
---

# BEMA agent instructions

## When to use BEMA

Use BEMA when the user asks to find a BEMA Podcast episode by title, topic, person, session, biblical passage, or episode number; compare where a theme appears across the podcast; retrieve structured episode, audio, chapter, or source metadata; locate transcript evidence; or hand back an episode page for listening.

Do not use BEMA for account management, data mutation, purchases, general web search, or claims about the podcast that are unsupported by episode data. BEMA is a discovery interface and not the podcast publisher. Source URLs in the data remain authoritative.

## Recommended workflow

1. For a known number or metadata question, call `GET https://api.bema.jpamorgan.com/v1/episodes?q={query}` and identify the stable `slug`; retrieve one full record with `GET /v1/episodes/{slug}`.
2. For a theme or phrase, search available transcript text with `GET https://api.bema.jpamorgan.com/v1/transcripts/search?q={query}&limit={1-30}`. Add `session={number}` only when the user requested a session.
3. Retrieve a selected transcript with `GET https://api.bema.jpamorgan.com/v1/transcripts/{slug}` before quoting, summarizing, or drawing a detailed conclusion.
4. Cite the episode page or transcript `sourceUrl`. Clearly distinguish quotations from summaries.
5. If transcript retrieval returns `404`, say that a local transcript is unavailable and rely only on catalog metadata or linked sources. Never synthesize missing transcript text.

## Access choices

For REST, follow the [OpenAPI specification](https://api.bema.jpamorgan.com/openapi.json) and [API guide](https://bema.jpamorgan.com/api.md). For native product tools, connect to the [BEMA product MCP endpoint](https://api.bema.jpamorgan.com/mcp). For documentation search, use the separate [documentation MCP endpoint](https://api.bema.jpamorgan.com/mcp/docs). Public catalog, transcript, NLWeb, A2A, and MCP operations are free, unauthenticated, and read-only. The separate `/agent/v1/session` demonstration is protected and stateful; follow [the authentication guide](https://bema.jpamorgan.com/auth.md).

For agent-to-agent episode discovery, send synchronous JSON-RPC requests to the [BEMA A2A v1.0 endpoint](https://api.bema.jpamorgan.com/a2a/v1). It accepts the canonical `SendMessage` method (and the `message/send` compatibility alias), requires no authentication, and advertises only the `search-bema-episodes` skill. It does not support streaming, tasks, push notifications, or transcript retrieval; use REST or MCP for those capabilities.

Machine discovery is available through the [Agent Skills index](https://bema.jpamorgan.com/.well-known/agent-skills/index.json), [A2A v1.0 agent card](https://bema.jpamorgan.com/.well-known/agent-card.json), [MCP service index](https://bema.jpamorgan.com/.well-known/mcp/index.json), [AI catalog](https://bema.jpamorgan.com/.well-known/ai-catalog.json), and [Schema Map](https://bema.jpamorgan.com/schemamap.xml). The [canonical source repository](https://github.com/jpamorgan/BEMA) is private and requires repository access; repository-only agent rules and source skills cannot be anonymously discovered there.