Model Context Protocol

PrompTessor MCP

Connect AI applications to PrompTessor tools, resources, and reusable prompt templates over Streamable HTTP.

Verified against the implementation ·

Server contract

PropertyValue
Endpointhttps://mcp.promptessor.com/mcp
TransportStreamable HTTP
Protocol revision2025-11-25 (with SDK-supported backward compatibility)
AuthenticationOAuth required
HTTP methodsGET, POST, DELETE, OPTIONS
Primitives33 tools, 3 resources, 2 prompts
Account accessVerified account with an active paid plan

API keys are REST-only

pt_live_ and legacy pt_test_ credentials are rejected by the MCP channel with oauth_required. Use the client’s OAuth flow.

How MCP maps to PrompTessor

  • Tools execute prompt workflows or read and manage account data.
  • Resources expose account usage, recent optimizations, and recent community prompts as application-controlled context.
  • Prompts provide two user-invoked templates for optimizing a current prompt and generating a ready-to-use prompt.
  • AI tools reserve the same product credit used by the web workspace and REST API.
  • Most AI tools save the same history as the equivalent product workflow; Reverse Prompt intentionally stores no history.

Client compatibility

ClientRemote HTTPOAuthToolsResourcesPromptsGuide reviewed
ChatGPT web pluginYesYesYesClient-dependentClient-dependent2026-09-09
ChatGPT desktop / CodexYesYesYesYesYes2026-09-09
Claude web / Desktop / mobileYesYesYesClient-dependentClient-dependent2026-09-09
Claude CodeYesYesYesYesYes2026-09-09
Gemini CLIYesYesYesClient-dependentClient-dependent2026-09-09
CursorYesYesYesClient-dependentClient-dependent2026-09-09
VS Code + GitHub CopilotYesYesYesClient-dependentClient-dependent2026-09-09
DevinYesYesYesClient-dependentClient-dependent2026-09-09

Protocol support

PrompTessor currently negotiates MCP 2025-11-25 and earlier revisions supported by the installed SDK: 2025-06-18, 2025-03-26, 2024-11-05, and 2024-10-07. It does not yet advertise the incompatible 2026-07-28 transport revision. Clients should use the protocol version returned by initialize.

Client-dependent means UI-dependent

The server registers all three MCP primitives. Some hosts currently expose tools while hiding resources or prompt templates; this is a host capability difference, not a failed PrompTessor connection.

Complete live verification before release

The date above records when each setup guide and client capability was reviewed, not a successful end-to-end production connection. Before launch, run the release test cases below against the deployed endpoint and record the client version and result.

First connection test

  1. 1

    Add the endpoint

    Configure https://mcp.promptessor.com/mcp as a remote Streamable HTTP server.

  2. 2

    Complete OAuth

    Sign in to PrompTessor and review the requested scopes. Clients that omit an explicit scope list receive all public scopes.

  3. 3

    List capabilities

    Confirm that the client discovers tools. Feature support differs: some clients show tools but not prompts or resources.

  4. 4

    Call get_usage

    This read-only tool validates authentication, scope, account access, and rate limiting without spending an AI credit.

Release and plugin-review test cases

Run these cases against the deployed endpoint and a dedicated paid review account before every public release or ChatGPT plugin submission. Record the actual tool chosen, output, credit change, history change, and approval behavior.

TypeUser requestExpected result
PositiveCheck my PrompTessor usage.Calls get_usage; returns plan, credits, and API limits; charges 0 credits.
PositiveCreate a ready-to-use launch prompt for ChatGPT.Calls generate_prompt with outputMode ready_to_use and a unique clientRequestId; creates generator history and charges 1 credit.
PositiveAnalyze and optimize this prompt into two versions.Calls analyze_and_optimize_prompt with variationCount 2; returns analysis and two versions.
PositiveList my latest five optimization history items.Calls list_history with type optimization and limit 5; no credit charge.
PositivePublish library item lib_review_1.Requests explicit confirmation, calls publish_library_item, and makes only that owned item public.
NegativeUse get_usage with a token lacking usage:read.Returns isError with insufficient_scope; no account data is returned.
NegativeDelete my prompt history. (No item identified.)Does not call a destructive tool; asks the user to select an exact type and id.
NegativeRetry generate_prompt with the same clientRequestId but different arguments.Returns idempotency_conflict and does not charge another credit.

Use a dedicated review account

The account must be verified, have an active eligible plan, avoid MFA or other login steps the reviewer cannot complete, and contain safe sample history and library items for the test cases.