# Claudforce — HLS — Provider + Payer

> An ordered, verified route from an empty Salesforce org to a working Data Cloud, Marketing Cloud Next and Agentforce demo, including every trap that cost a day. Written from the org it was built in and verified against org trailsignup-a612cacbdcac92 on 3 September 2026.

A 200 is not a result. Every write in this document is verified with a read-back query, and every failure mode is labelled trap, broken or verified.

## Read this first

- [Full guide, Markdown](https://claudforce-hls.thelonelyagent.com/rebuild-path.md): the entire document in one file.
- [Full guide, plain text](https://claudforce-hls.thelonelyagent.com/rebuild-path.txt): identical content, for clients that cannot read the .md content type.

## Live MCP server

- Endpoint: https://claudforce-hls.thelonelyagent.com/api/public/mcp (Streamable HTTP, no auth)
- Tools (twelve): start_here, get_guide, get_corrections, list_sections, get_section, search_traps, verify_step, list_skills, get_skill, get_dataset_schema, get_manifest, list_downloads
- start_here is the first call. It states the two-server setup and the order to work in.
- get_guide returns an overview and the section index by default; pass part: "full" for the entire document.
- get_corrections returns the dated corrections table: what this guide got wrong and when it was fixed.
- search_traps takes an optional limit, default 20, and always reports the total matched.
- verify_step returns the check for one step: what to run against the target org, what a correct result looks like, what a wrong one means, and what the step does not prove.
- list_skills and get_skill return the six Claude skills in full, as Markdown, including the claudforce-hls-rebuild orchestration skill.
- get_dataset_schema returns the shape of all thirteen mock datasets: file, row count, columns with types, and join keys. It never returns the CSV bytes.
- get_manifest returns the full file list of one pack: asset-pack, mock-data or skills-pack.

Claude Code: `claude mcp add --transport http claudforce-hls https://claudforce-hls.thelonelyagent.com/api/public/mcp`

## You need two servers, not one

This server carries the knowledge: the ordered route, the verified traps, the skills, the dataset
shapes and the anchor resolvers. It has no access to any Salesforce org and never will.

To build anything you also need the Salesforce MCP server connected to your own org. That is what
issues the calls. Connect both, then work through the foundation section in order.

Connecting the Salesforce org requires an External Client App, not a Connected App, with the
`mcp_api` and `refresh_token` scopes and the callback `https://claude.ai/api/mcp/auth_callback`. The
app can take up to thirty minutes to propagate, so a first-attempt failure is usually propagation
rather than misconfiguration.

## Downloads

- [Salesforce skills pack](https://claudforce-hls.thelonelyagent.com/salesforce-skills-pack.zip)
- [Mock data](https://claudforce-hls.thelonelyagent.com/makana-mock-data.zip)
- [Rebuild path asset pack](https://claudforce-hls.thelonelyagent.com/rebuild-path-asset-pack.zip)

## Manifests

Every file in each zip, one per line, with byte size. Read these before downloading.

- [Asset pack manifest](https://claudforce-hls.thelonelyagent.com/manifests/asset-pack.txt)
- [Mock data manifest](https://claudforce-hls.thelonelyagent.com/manifests/mock-data.txt)
- [Skills pack manifest](https://claudforce-hls.thelonelyagent.com/manifests/skills-pack.txt)

## Sections

- This page is an MCP server. Point your agent at it.
- The downloads
- What the build produces
- What you are rebuilding
- Before you build anything
- The data foundation
- Making it portable
- The identity spine
- Consent, then segments
- Engagement, campaigns, content
- Dashboards
- Custom components
- Agents
- Customer evidence
- Verification
- Traps index
- Corrections
- Downloads