Config Connector (MCP)

Manage a brand’s response-automation config from Claude, through the engine API’s built-in MCP server.

Config Connector (MCP)

Manage a brand’s response-automation config from Claude. The MCP server is the already-running engine API — there is no separate process to deploy.

Server name	stateset-config-connector
HTTP endpoint	POST https://api.workstream.stateset.com/v1/mcp
stdio (local)	engine-api mcp-stdio
Protocol	MCP 2025-06-18, JSON-RPC over Streamable HTTP
Auth	Authorization: Bearer <api-key>

The transport is POST-only; GET returns 405 and compliant clients fall back to POST.

Auth accepts either a per-brand API key, which scopes Claude to that brand’s tenant, or a global admin key.

Tools

Tool	Purpose
list_brands	Brands visible to the caller — id, slug, status, config version
get_brand_config	Read a brand’s current deterministic_config and version
get_brand_scorecard	Closed-loop quality rates (approval, automation, escalation) so edits are grounded in real performance
list_config_versions	Published config history, for audit
propose_config_patch	Apply JSON-Pointer edits and return the result without saving
validate_config	Validate a proposed or current config
apply_config	Commit a new config through the audited, versioned pipeline

apply_config requires a change_note and supports expected_config_version for optimistic concurrency — so two people editing the same brand can’t silently overwrite each other.

The natural loop

get_brand_config + get_brand_scorecard
        ↓
propose_config_patch      (nothing saved yet)
        ↓
validate_config
        ↓
apply_config              (audited, versioned)

get_brand_scorecard is what makes this more than a text editor. Reading approval and escalation rates before proposing a change means the model is tuning against observed behavior rather than guessing at a threshold.

1. Mint a per-brand API key

curl -X POST https://api.workstream.stateset.com/v1/brands/<BRAND_ID>/api-keys \
  -H "Authorization: Bearer <ADMIN_KEY>" \
  -H "Content-Type: application/json" \
  -d '{"label":"claude-mcp"}'

The plaintext token is returned once. Store it securely — it cannot be retrieved later, only revoked and reissued.

2. Connect from Claude

Claude Code

claude mcp add stateset-config \
  --transport http https://api.workstream.stateset.com/v1/mcp \
  --header "Authorization: Bearer <BRAND_API_KEY>"

Then in a session:

List brands, read that brand’s config and scorecard, and propose lowering the review-gate confidence to 0.7.

Claude Desktop

Bridge the remote HTTP server to stdio:

{
  "mcpServers": {
    "stateset-config": {
      "command": "npx",
      "args": [
        "-y", "mcp-remote",
        "https://api.workstream.stateset.com/v1/mcp",
        "--header", "Authorization: Bearer <BRAND_API_KEY>"
      ]
    }
  }
}

Scoping

Prefer a per-brand key over an admin key. The brand key confines every tool call to that brand’s tenant, so a mistaken apply_config cannot reach another customer’s configuration.

Related

• Control plane — the versioned config pipeline behind apply_config
• Configuration — what lives in deterministic_config
• Workflow Studio — the UI equivalent
• All MCP servers