Skip to main content

​
Overview

ApiTraffic exposes a built-in MCP (Model Context Protocol) server that allows AI assistants to securely query your API traffic data. This enables AI-powered debugging, analytics, and monitoring workflows directly from tools like Claude Desktop, Cursor, Windsurf, VS Code, and any other MCP-compatible client.
MCP tokens are scoped to a specific environment and set of buckets — AI assistants can only access the data you explicitly allow.

​
Available Tools

When connected, the MCP server exposes the following tools to AI assistants:

​
Request & Metrics Tools

ToolDescription
search_requestsSearch API request logs with filtering by method, status code, time range, and query string
get_metricsGet summary metrics: total requests, average response time, error count, and blocked count
get_errorsList requests that returned 4xx or 5xx status codes
get_request_detailGet full details for a single request including headers, body, and response
get_throughputGet request throughput time series broken down by status code range (2xx, 3xx, 4xx, 5xx)
get_response_timesGet response time percentiles (p50, p95, p99) and averages over time

​
Configuration Tools

ToolDescription
list_bucketsList the buckets the MCP token has access to
list_environmentsList environments configured for the account (e.g. production, staging, development)
list_redactionsList data redaction rules that mask sensitive data from captured requests
list_exclusionsList exclusion rules that prevent matching requests from being captured

​
Event Tools

ToolDescription
list_eventsList event markers (deployments, incidents, config changes) to correlate with traffic data
create_eventCreate an event marker to correlate with API traffic (e.g. deployments, incidents)

​
Quick Start

​
1. Create an MCP Token

Navigate to Settings → MCP Tokens in the ApiTraffic dashboard and create a new token. Select the environment and buckets you want the AI assistant to access.
The token value is only shown once at creation. Copy it immediately and store it securely.

​
2. Configure Your AI Client

{
  "mcpServers": {
    "apitraffic": {
      "serverUrl": "https://api.apitraffic.io/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_MCP_TOKEN"
      }
    }
  }
}
Config file locations:
  • Windsurf: ~/.codeium/windsurf/mcp_config.json
  • Cursor: ~/.cursor/mcp.json (global) or .cursor/mcp.json (project)
  • Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS)
  • VS Code: .vscode/mcp.json (project)

​
3. Start Querying

Once connected, you can ask your AI assistant questions like:
  • “Show me the most recent API errors in production”
  • “What’s the average response time for the last hour?”
  • “Search for all POST requests to /api/users that returned 500”
  • “Give me throughput metrics for the past 24 hours”
  • “List my environments and redaction rules”
  • “Create a deployment event for v2.3.1”

​
Transport Protocols

ApiTraffic supports two MCP transport protocols: The primary transport with stateful session management. Sessions persist across requests using the mcp-session-id header and are backed by Redis for multi-instance reliability.
  • Endpoint: POST https://api.apitraffic.io/mcp
  • Auth: Authorization: Bearer YOUR_MCP_TOKEN header
  • Session: Automatic via mcp-session-id response header

​
SSE (Server-Sent Events)

Legacy transport maintained for backward compatibility with older MCP clients.
  • Endpoint: GET https://api.apitraffic.io/mcp/sse
  • Message Endpoint: POST https://api.apitraffic.io/mcp/sse/message
  • Auth: Authorization: Bearer YOUR_MCP_TOKEN header

​
Security

  • Scoped access — MCP tokens are read-only for traffic data; the only write operation is create_event
  • Environment scoped — Each token is locked to a single environment
  • Bucket scoped — Tokens can be restricted to specific buckets or granted access to all buckets in the environment
  • Instant revocation — Revoking a token immediately terminates access
  • No cross-tenant leakage — Token context is enforced on every tool call
  • Session isolation — Each MCP session is isolated with its own server instance

​
Example Interactions

​
Searching for Errors

Ask your AI assistant:
“Find all 500 errors in the last 2 hours and show me the request details”
The assistant will use get_errors to find matching requests, then get_request_detail to retrieve full request/response bodies for debugging.

​
Monitoring Performance

“What are the p95 response times for my API over the last day? Are there any anomalies?”
The assistant will use get_response_times and get_throughput to analyze performance patterns and identify spikes.

​
Investigating Issues

“Search for all requests from user-agent containing ‘bot’ that returned 429 status codes”
The assistant will use search_requests with the appropriate filters to find rate-limited bot traffic.

​
Tracking Deployments

“Create a deployment event for version 2.3.1 released from GitHub Actions”
The assistant will use create_event to create a marker that correlates with your traffic metrics, making it easy to spot regressions.

​
Reviewing Configuration

“Show me all my redaction rules and exclusions”
The assistant will use list_redactions and list_exclusions to display your data protection and traffic filtering configuration.