AI integrations

Bring customer-aware analytics to your AI client.

A remote, project-scoped MCP server exposes deterministic Apirelio analytics through eight read-only tools.

Read-onlyProduction ready
01

Boundary

Analytics, not an autonomous agent.

Apirelio's MCP server adapts existing analytics services to Model Context Protocol. It does not call OpenAI or another language model, does not need a provider API key and cannot modify project data.

R

Read-only

Every tool is declared read-only and exposes bounded aggregate results, never raw telemetry events.

1

One project

Each token belongs to exactly one project and cannot request another tenant.

60

Bounded

Calls are rate limited, cached briefly and capped by list and response-size limits.

02

Connect

Create and configure a connection.

Open AI & MCP in project navigation. A workspace owner names the connection, chooses 30, 90, 180 or 365 days, then copies the project server URL and one-time bearer token into the MCP client.

Streamable HTTP initializehttp
POST /mcp/projects/YOUR_PROJECT_ID
Authorization: Bearer apr_mcp_••••••••
Content-Type: application/json
Accept: application/json, text/event-stream

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "initialize",
  "params": {
    "protocolVersion": "2025-06-18",
    "capabilities": {},
    "clientInfo": { "name": "Your client", "version": "1.0" }
  }
}
The secret is shown once

Apirelio stores only a SHA-256 hash. If the token is lost, revoke the connection and create another one.

03

Client setup

Connect Claude Code or OpenAI Codex.

Apirelio uses the standard Streamable HTTP transport. In any compatible MCP client, enter the project server URL and send the connection token as an Authorization: Bearer header. Do not select stdio or the deprecated SSE transport.

Replace YOUR_PROJECT_ID before connecting

YOUR_PROJECT_ID is a placeholder, not a valid value. Replace it with your numeric Apirelio project ID. For example, if the AI & MCP page shows https://apirelio.com/mcp/projects/17, replace YOUR_PROJECT_ID with 17. The safest option is to copy the complete Server URL directly from that page.

Claude Code

Export the token as APIRELIO_MCP_TOKEN, save this as .mcp.json, then run claude mcp list or open /mcp. Keep the token out of the JSON file.

.mcp.jsonjson
{
  "mcpServers": {
    "apirelio": {
      "type": "http",
      "url": "https://apirelio.com/mcp/projects/YOUR_PROJECT_ID",
      "headers": {
        "Authorization": "Bearer ${APIRELIO_MCP_TOKEN}"
      }
    }
  }
}

OpenAI Codex CLI, IDE and desktop

Export APIRELIO_MCP_TOKEN and add this table to ~/.codex/config.toml. Codex CLI, its IDE extension and the ChatGPT desktop app share this configuration. Verify it with codex mcp list or /mcp.

~/.codex/config.tomltoml
[mcp_servers.apirelio]
url = "https://apirelio.com/mcp/projects/YOUR_PROJECT_ID"
bearer_token_env_var = "APIRELIO_MCP_TOKEN"
startup_timeout_sec = 30
tool_timeout_sec = 60
enabled = true
Using multiple projects

Create a separate connection in each project and add each one to your client under a descriptive name such as apirelio-orders or apirelio-billing. Each server entry uses that project's URL and token, so analytics and credentials remain isolated.

04

Example session

From prompt to evidence in seconds.

This anonymized OpenAI Codex session shows two typical workflows. Codex selects explicit Apirelio tools, receives bounded structured analytics and turns the returned metrics into an evidence-based answer.

Animated OpenAI Codex session calling Apirelio customer health and release impact MCP tools
Illustrative sample data. Apirelio supplies deterministic analytics; Codex writes the final answer. Open the animation for a full-size view.
05

Capabilities

Eight explicit analytics tools.

get_project

Safe project context and connection status.

get_attention_items

Prioritized critical and warning items.

search_customers

Resolve a customer name to a stable ID.

get_customer_health

Explain one customer health score and risks.

investigate_incident

Aggregate diagnosis for a stable error code.

compare_release

Compare a release with its baseline.

summarize_changes

Compare a period with the preceding period.

get_integration_journey

Integration funnel and stalled customers.

There is no general SQL, HTTP, URL-fetch or arbitrary analytics-query tool. Default periods cover the last seven days and cannot exceed 90 days.

06

Ask

Questions grounded in evidence.

  • Which customers need attention this week, and why?
  • Did release 2026.08.4 improve reliability?
  • Investigate error code PAYMENT_TIMEOUT in production.
  • Which integrations are stalled before regular usage?

Tool summaries are deterministic. Your MCP client decides how to combine the returned metrics into its answer.

07

Responsibility

Keep project credentials scoped.

Store the bearer token in your MCP client's secret storage, never source control or a URL. Revoke unused connections immediately and prefer the shortest practical expiration.

External AI data handling

Analytics sent to an AI client are subject to that client's processing, privacy and retention rules. Review them before connecting production projects.

08

Recover

Authentication and limits.

UNAUTHENTICATED means the token is missing, invalid, expired, revoked, lacks analytics:read or belongs to another project. Create a replacement connection if the secret cannot be recovered.

A rate-limited response means the connection exceeded its per-minute allowance. Retry after the indicated window. For oversized or slow analytics, shorten the period or lower the requested list limit.