MCP Server · v2 · OAuth

APIWORX MCP Server

Give AI assistants read-only access to your APIWORX operations data over the Model Context Protocol. Assistants can see which flows exist, how many runs happened, success vs. failure, records processed, and which errors are occurring — always scoped to your customer.

Endpoint

https://app.apiworx.com/mcp

Transport: MCP Streamable HTTP · Auth: OAuth 2.1 sign-in with your APIWORX account

Quickstart

In your assistant, add a custom MCP connector with the URL https://app.apiworx.com/mcp.
The assistant opens an APIWORX sign-in page. Sign in and approve access.
Ask “describe the APIWORX data”, then “show my run metrics by day for the last 30 days”.

Authentication

The connector uses OAuth: no tokens to copy or paste. The assistant acts as the APIWORX user who approved it, with that user's permissions. You can revoke access from the assistant at any time.

Claude

Settings → Connectors → Add custom connector → paste https://app.apiworx.com/mcp → Connect.

Cursor / other clients

{
  "mcpServers": {
    "apiworx": { "url": "https://app.apiworx.com/mcp" }
  }
}

Clients that support MCP OAuth will prompt you to sign in on first use.

Available tools

Tool	Purpose	Parameters
echo	Echo text back — verifies connectivity.	text (required)
whoami	Return the signed-in user's id and email from the OAuth token.	—
describe_data	Explains data sources, every status value and metric, access rules and limits. Call this first.	—
list_customers	Customers (tenants) with identifiers for matching to other systems: customer_id, esb_org_id, hubspot_company_id, email_domain, status, flow count.	limit (default 100, max 500), cursor
list_flows	Flows with source/destination connector, schedule, status (active \| paused \| draft \| errored \| deleted \| unknown), tags, last run.	customer_id, status, limit, cursor
get_flow_runs	Individual executions, newest first: timing, status, record counts, error summary.	customer_id and/or flow_id, date_from, date_to, status (Successful \| Error \| InProgress), limit, cursor
get_run_metrics	Dashboard feed aggregated in the database: total, succeeded, failed, success rate, records processed, avg duration.	customer_id, date_from, date_to, group_by (day \| week \| flow \| customer)
get_errors	Grouped error patterns with count, first/last seen, affected flows and triage status.	customer_id, date_from, date_to, limit
get_customer_summary	One-call snapshot for a customer: identity, flows by status, run totals, per-flow metrics, top 10 errors.	customer_id, date_from, date_to

Dates are ISO 8601 (UTC). Windows default to the last 30 days (max 400). List tools return up to 100 rows by default (max 500) plus a next_cursor for the next page.

Read-only & safety

Every tool is read-only — none can create, change, delete, or start a flow or run.
No secrets are returned: no connector credentials, tokens, API keys, or AWS identifiers, and no record contents — only counts and error messages.
Every call is recorded in an audit log (user, tool, parameters, time).

Tenant scoping

Regular users can only read their own customer. Asking for another customer returns Forbidden.
APIWORX internal admins can read all customers (omit customer_id) or pick one.
Scoping is enforced on the server and again by database access rules.
customer_id is the join key to other systems, alongside esb_org_id, hubspot_company_id, and email_domain.

Typical queries

“Success rate per customer this week.” → get_run_metrics (group_by: customer)
“Which Vector flows are paused?” → list_flows
“Top errors for APIWORX in the last 7 days.” → get_errors
“Give me a one-page summary of Metis for September.” → get_customer_summary