Working with the Sales Platform

MyChatBot has two separate runtimes that work together. Claude or Codex normally configures both during prototyping; a validated routine then becomes the durable orchestrator around Sales processes.

	Sales Platform	Agents Platform
Role	Customer-conversation and CRM runtime	Production orchestration runtime
What it does	Holds channel conversations, runs sales assistants and native follow-ups, and owns clients, leads, pipeline state, and orders	Runs routines for batch processing, reviews, enrichment, reporting, and cross-system coordination through agents and tools
Who it talks to	Your customers	You, external systems, and—through Sales tools—the Sales Platform

The short version: the Sales Platform remains the runtime for customer conversations and its native follow-ups. What moves to the Agents Platform is higher-level orchestration around those sales processes. A routine can inspect or change Sales records through an enabled agent, but customer replies stay in the Sales chat rather than creating a new Agents conversation.

See the complete Agents Platform operating model for the configuration-to-production flow.

Configure both platforms with Claude or Codex

For a non-trivial system, connect both account-scoped MCP surfaces to the same Claude Code or Codex session:

During configuration	Use it for
Agents MCP	Agents, skills, connectors, Business Knowledge, routine YAML, validation, bounded preview, schedules, and supported triggers
Sales-Management MCP	Sales assistants, clients, chats, channels, CRM/pipeline resources, knowledge, outreach, and other Sales primitives

The coding agent inspects and prototypes in those MCP surfaces. At production runtime, the routine does not depend on the external Claude/Codex session: its leaf steps execute on MyChatBot agents, and those agents reach Sales through their enabled Sales Platform connector or tool.

The three ways they connect

Connection	What it lets an agent do	Turn it on
Sales Platform capability	Operate your Sales Platform — create & configure sales assistants, connect channels, build pipelines, manage leads, trigger outreach, read chats	Enable the Sales Platform tool on an agent (the Sales Platform Wizard has it by default)
Sales-Platform knowledge as agent knowledge	Read your live FAQ, product catalog, feed, spreadsheet, and business documents — the same content your sales assistant uses	Attach them under the agent's Knowledge → From your sales platform, then turn on Business knowledge
Shared product search	Look up products from the same catalog your sales assistant searches	Attach the product source as knowledge (above) — see How search works

Start with Claude or Codex for a complete system

Use the MCP surface map to connect the Agents and Sales-Management surfaces, then describe the business outcome rather than individual clicks. The built-in Sales Platform Wizard remains useful for a quick one-off setup or repair inside MyChatBot, but Claude/Codex is the primary workspace for designing and dry-running the whole system.

Scenario 1 — Stand up a sales assistant from Claude or Codex

Describe the customer journey once and let the coding agent configure the Sales Platform through its management MCP.

1. Connect a focused Sales-Management MCP area—or the combined surface when the setup genuinely crosses several areas.
2. Ask Claude or Codex to inspect the existing assistants, channels, pipeline, and knowledge before proposing changes.
3. Review the proposed assistant instructions, channel assignments, product/FAQ sources, and pipeline stages before approving writes.
4. Provide credentials or complete OAuth in the MyChatBot app; never paste secrets into a prompt or routine.
5. Test with controlled data and confirm the customer conversation stays on the Sales Platform.

You now have a customer-facing Sales runtime that a later Agents routine can inspect and orchestrate around.

Scenario 2 — Kick off outreach, hand the conversation over

This is the classic hand-off: an agent starts the conversation, the Sales Platform runs it.

1. From a routine step whose executing agent has the required Sales Platform capability, deliberately trigger a one-off outreach message to a lead—or preview an audience and send a controlled campaign.
2. The sales assistant sends the first message on the channel.
3. The customer replies — and the sales assistant takes over: it answers questions from your catalog and FAQ, qualifies the lead, and moves it along the pipeline.
4. If the lead goes quiet, a follow-up automation chases them — e.g. "if a lead in the Contacted stage hasn't replied in 3 days, send a check-in." Configure this as a native Sales follow-up through the Sales-Management MCP or the Sales app.

You triggered one message; the Sales Platform handled the entire back-and-forth and the follow-ups on its own.

Where the reply shows up

The customer's reply and the whole conversation live on the Sales Platform (in Leads / Active chats), on whatever channel they came in on — not in the agent's chat. The agent's job was to start it and, if you like, report on it later (Scenario 4).

Scenario 3 — One source of product truth

Build your product catalog and FAQ once, on the Sales Platform, and use it everywhere.

• Your customer-facing sales assistant already searches it.
• Attach the same sources to your Agents-Platform agents (Knowledge → From your sales platform) so an internal research or content agent answers from the exact same live catalog — no second copy to maintain.

Update the catalog once and every agent, on both platforms, reflects it on the next answer. See Knowledge bases and How search works.

Scenario 4 — A scheduled ops digest

Use a scheduled routine to poll bounded Sales changes and produce a morning review.

1. Give the routine's executing agent the required Sales read capabilities and run the routine on a schedule (for example, weekdays at 09:00).
2. Read clients or chats in bounded pages changed since a stored timestamp, cursor, or stable high-water mark. Keep the total-record loop in the routine or a skill script.
3. Use bounded foreach concurrency to summarize conversations, review pipeline state, and flag quiet leads without one unbounded fetch.
4. Store processed identifiers and advance the cursor only after a page completes, so a retry does not duplicate an effect.
5. Post the digest to the routine conversation; keep follow-ups draft-only until a narrow live check proves the write path.

The Wizard's Summarize recent customer chats, Review my pipelines and stages, and Find leads matching a filter tasks are useful one-off prototypes. Promote the combined repeatable process into the scheduled routine described above.

Built-in routines are useful starting points

The Wizard ships with ready-made multi-step routines that cover this ground in one click: Weekly business review (conversations, leads, pipeline → a short report with three recommendations) and Lead follow-up review (finds quiet leads and drafts a follow-up for each — nothing is sent without you). Find them on the Wizard's Tasks tab, under Routines.

Current feedback-loop boundary

Data and actions can flow from an Agents routine into the Sales Platform through Sales tools. Native Sales events do not currently trigger an Agents routine. A new Sales chat, customer reply, lead-stage change, or Sales order will not start one automatically.

Use a scheduled polling routine for state that exists only in Sales. Use an app-event trigger for supported external apps such as Stripe, Gmail, GitHub, or Slack, or a secret webhook when an outside service can call it. This avoids creating one Agents conversation per customer chat: the scheduled routine keeps its own durable conversation and processes many Sales records in bounded batches.

Which platform for which job

Rule of thumb

• Ongoing customer conversations, across channels, day and night → Sales Platform.
• Configuration and prototyping across both platforms → Claude or Codex through MCP.
• Repeatable batch work, reporting, enrichment, and cross-system coordination → an Agents Platform routine (whose agents can reach into Sales).
• One-off work, inspection, approvals, and debugging → an Agents conversation.
• Product & FAQ knowledge → build it once on the Sales Platform, reuse on both.

See also

• Agents Platform operating model — routine-first ownership, promotion path, and feedback-loop boundary
• Build routines with Claude or Codex — configure, validate, preview, and stage a routine
• MCP surface map — choose the Agents and Sales configuration surfaces
• Custom connectors — the Sales Platform capability and the per-domain connectors an agent can use
• Knowledge bases — attaching your Sales-Platform FAQ/catalog as agent knowledge
• Tasks & schedules — run a routine on a deliberate cadence
• Routines — durable orchestration for repeatable work
• Best practices & agent design — keep each agent focused; use delegation and teams
• Sales Platform — the customer-facing side