Skip to Content
MCP

OTseek MCP Server Documentation

Description

OTseek exposes a hosted MCP server for benchmarking AI agent checkout performance and for submitting new website benchmark requests. The connector is named OTseek, the production endpoint is https://backend.otseek.com/mcp, and the server uses streamable HTTP transport.

The current OTseek MCP server is focused on two workflows:

  • analyzing OTseek’s hotel checkout benchmark dataset through a single read-only query tool
  • collecting contact information for organizations that want OTseek to benchmark their own website

Features

Hosted MCP Endpoint

  • Remote MCP endpoint at https://backend.otseek.com/mcp
  • Streamable HTTP transport
  • No OAuth flow or OTseek account connection required today
  • Backend-enforced transport security and MCP-specific security headers

Benchmark Analysis

  • Structured hotel checkout benchmark data across OTseek and major travel sites
  • Support for raw row inspection plus computed rank, list, and count results
  • Filtering by model, company, checkout status, failure reason, stage reachability, and upsell handling
  • An HTML benchmark widget for MCP clients that support app-style tool output rendering

Website Benchmark Intake

  • Built-in write tool for benchmark intake
  • Validates name, email, company name, site URL, and optional notes
  • Stores the request so OTseek can follow up by email
  • Rejects malformed input and enforces note length limits

Available Tools

The following tools are currently available through the OTseek MCP server:

query-hotel-checkout-performance

Read-only benchmark analysis tool for OTseek’s hotel checkout dataset.

It supports four operations:

  • rows to return matching benchmark rows
  • rank to compute the fastest or slowest result for a metric
  • list to list matching companies or models
  • count to count matching rows, companies, or models

Common filters:

  • model: gpt-5.4 or opus-4.6
  • companies: OTseek, Booking.com, Priceline, Hotels.com, Hopper, Expedia, Reservations.com, HotelPlanner, Vio, Traveluro
  • status: successful or failed
  • failureReason: captcha, popup, or redirect
  • reachedStage with hasReachedStage
  • upsellHandling: consulted_user, auto_declined, or na

submit-website-benchmark-request

Write tool for organizations that want OTseek to benchmark their own website.

Required fields:

  • name
  • email
  • company_name
  • site_url

Optional field:

  • request_notes up to 500 characters

Available Resources

The following MCP resource is currently exposed by the OTseek MCP server:

hotel-checkout-widget

OTseek also exposes an HTML widget resource used to render hotel checkout benchmark results inside MCP clients that support interactive tool output.

Connector Setup

Hosted Connector

If your MCP client provides a connector directory or remote-server browser:

  1. Search for OTseek.
  2. Select the OTseek connector.
  3. Connect the server.
  4. Review any requested permissions.

No OAuth login is currently required.

Manual Remote MCP Setup

If your client supports direct remote MCP configuration, use:

  • Connector name: OTseek
  • URL: https://backend.otseek.com/mcp
  • Transport: streamable HTTP
  • Authentication: none

A common remote-MCP configuration looks like this:

{ "mcpServers": { "OTseek": { "url": "https://backend.otseek.com/mcp" } } }

Common Use Cases

Example 1: Inspecting the Full Benchmark

User prompt: “Show me the hotel checkout benchmark rows for gpt-5.4.”

What happens:

  • OTseek returns the matching benchmark rows for gpt-5.4
  • The response includes structured timing fields and display-ready durations
  • MCP clients with widget support can render the benchmark output visually

Example 2: Ranking Fastest Checkout Performance

User prompt: “Which company had the fastest successful checkout overall?”

What happens:

  • OTseek runs the rank operation with metric=total and direction=fastest
  • The response identifies the winning company and the relevant duration
  • The output can be narrowed further by model or company filters

Example 3: Filtering Failures and Upsell Handling

User prompt: “How many failed checkouts hit a captcha?”
User prompt: “List companies that auto-decline upsells.”

What happens:

  • OTseek uses count or list on the benchmark dataset
  • The tool applies filters such as status=failed, failureReason=captcha, or upsellHandling=auto_declined
  • The response returns a direct answer plus structured data for follow-up questions

Example 4: Submitting a New Benchmark Request

User prompt: “I want OTseek to benchmark our booking flow at https://example.com/checkout .”

What happens:

  • OTseek collects the required contact and company details
  • The MCP server validates the URL and request payload
  • The benchmark request is stored for OTseek follow-up

Change Log

Version 1.0 - 03 April 2026

  • Initial OTseek MCP server launch
  • Documents the hosted streamable HTTP endpoint
  • Documents the hotel checkout benchmark tool, benchmark intake tool, and widget resource

Troubleshooting

  • If your MCP client cannot connect, confirm the server URL is exactly https://backend.otseek.com/mcp and that your client is using remote MCP over streamable HTTP.
  • If the connector does not appear in a client directory, configure OTseek manually using the production MCP endpoint instead of directory search.
  • If browser-based clients fail to connect, verify that your client is allowed to reach the public HTTPS endpoint and that any network policy or firewall allows Claude or ChatGPT remote MCP traffic.
  • If a benchmark query returns no useful result, check that the requested filters are valid and compatible, especially operation, status, failureReason, reachedStage, and hasReachedStage.
  • If a benchmark request submission fails, confirm that name, email, company_name, and site_url are present and valid, and keep request_notes within the 500-character limit.

Support & Resources

Getting Help

Privacy & Data

OTseek’s MCP server:

  • returns precomputed benchmark data for query-hotel-checkout-performance
  • stores submitted contact details for submit-website-benchmark-request so OTseek can follow up by email
  • does not require OAuth login for the current connector flow

See also: