APIXX MCP Server

Expose your APIXX workspace to AI assistants over the Model Context Protocol. Once connected, Claude / Cursor / ChatGPT can list flows, debug runs, inspect field mappings, and trigger on-demand syncs — all scoped to your tenant.

Quickstart

1. Go to Settings → MCP Server and click New token. Copy the mcp_… secret — it's shown once.
2. Drop the snippet for your assistant below into its config file (or paste the URL+token into its UI).
3. Restart the assistant and ask “list my apixx flows”. You should see live data.

Tip: issue one token per assistant / device so revocation stays surgical, and set an expiry for short-lived experiments.

Authentication

Generate dedicated MCP tokens from Settings → MCP Server. Tokens are prefixed mcp_, shown once at creation, and scoped to the customer of the user that issued them. They are independent from your REST API keys — revoke them separately.

POST https://app.apixx.io/mcp
Authorization: Bearer mcp_••••••••
Accept: application/json, text/event-stream
Content-Type: application/json

• Use one token per assistant / device so revocation is surgical.
• Set an expiry for short-lived experiments (30 / 90 / 365 days).
• If a token leaks, revoke it in Settings → MCP Server — calls fail immediately.

Client configuration

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "apixx": {
      "url": "https://app.apixx.io/mcp",
      "headers": { "Authorization": "Bearer mcp_••••••••" }
    }
  }
}

Cursor

Open Settings → MCP → Add new MCP server, or edit ~/.cursor/mcp.json:

{
  "mcpServers": {
    "apixx": {
      "url": "https://app.apixx.io/mcp",
      "headers": { "Authorization": "Bearer mcp_••••••••" }
    }
  }
}

ChatGPT (Custom Connector)

In ChatGPT, open Settings → Connectors → Add custom connector and provide:

• Server URL: https://app.apixx.io/mcp
• Auth: Bearer token → paste your mcp_… secret

Custom connectors require a paid ChatGPT plan. Once enabled, the model can call any APIXX tool via natural language.

OpenAI Agents SDK

from agents import Agent
from agents.mcp import MCPServerStreamableHttp

server = MCPServerStreamableHttp(
    name="apixx",
    params={
        "url": "https://app.apixx.io/mcp",
        "headers": {"Authorization": "Bearer mcp_••••••••"},
    },
)

agent = Agent(name="apixx-ops", mcp_servers=[server])

Probing manually

curl -X POST https://app.apixx.io/mcp \
  -H "Authorization: Bearer mcp_••••••••" \
  -H "Accept: application/json, text/event-stream" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Available tools

All tools are read-only except trigger_sync, and all calls are tenant-scoped.

Typical queries

These are the kinds of prompts that work well once APIXX is wired into your assistant. Each one maps to one or more tools above — the model picks the right combination.

Daily operations

• “Which flows ran in the last 24 hours and how many records did each process?” → list_runs
• “Show me every flow that's currently failing or degraded.” → list_flows + get_system_health
• “Give me a one-line status summary for each flow.” → list_flows
• “What was the last successful run of the Salesforce → Snowflake accounts flow?” → list_flow_runs
• “Summarize today's system health and call out anything that needs attention.” → get_system_health

Debugging a failed run

• “The HubSpot contacts flow failed this morning — what went wrong?” → list_flow_runs + get_run_errors
• “Group the errors on run run_abc123 by error type and show a sample payload for each.” → get_run_errors
• “What are the top 5 error classes on the Shopify orders flow over the last week?” → get_flow_error_summary
• “Find every run in the last hour with status FAILED and tell me which connector they share.” → list_runs + list_connectors
• “Pull the full detail for run run_abc123, including timing and record counts.” → get_run

Understanding a flow

• “What source and destination does the NetSuite invoices flow use?” → get_flow
• “Show me the field mappings for the HubSpot → Salesforce contacts flow.” → get_flow_mappings
• “Which fields in the orders flow have filters or validations applied?” → get_flow_mappings
• “Does any flow write to the customers table in Snowflake?” → list_flows + get_flow_mappings

Connectors & health

• “List every connector I have configured and its current auth status.” → list_connectors
• “Are any of my connectors close to their rate limit or quota?” → list_connectors + get_system_health
• “What's my overall error rate across all flows this week?” → get_system_health + list_runs

Taking action

• “Trigger a sync for the Salesforce accounts flow and give me the run id.” → trigger_sync
• “Re-run every flow that failed in the last hour.” → list_runs + trigger_sync
• “Kick off the nightly NetSuite invoice sync now and watch it until it finishes.” → trigger_sync + get_run

Reporting & standups

• “Write a Slack-ready summary of integration health for this morning's standup.”
• “Draft a postmortem for yesterday's failed Shopify sync using the run + error data.”
• “How many records did we move across all flows yesterday, broken down by destination?”

Scopes & safety

• Every tool resolves the caller's customer_id from the bearer token. There is no cross-tenant access.
• trigger_sync is the only write operation; everything else is read-only.
• The server respects the same rate limits and quota caps as your REST API key.
• Token usage is tracked — see last_used_at in Settings → MCP Server.

Errors

MCP errors follow JSON-RPC. Common cases:

See also

• REST API reference — same data, programmatic access.
• Webhooks — push events instead of polling.
• Settings → MCP Server — create & manage tokens.