MCP reference

MCP resources and prompts

Use PrompTessor context resources and user-invoked prompt templates when the MCP client supports them.

Verified against the implementation ·

Resources

Resource URIScopeContents
promptessor://account/usageusage:readUnified account usage and public API policy
promptessor://history/optimizations/recenthistory:readThe 10 most recent optimization history items
promptessor://library/community/recentlibrary:readThe first page of recent Community Prompts

Resources return application/json text. The host decides how and when to attach resource contents to model context.

{
  "jsonrpc": "2.0",
  "id": 10,
  "method": "resources/list",
  "params": {}
}

Authorization and limits

Reading a resource requires the scope listed above and consumes one read-rate-limit request, but no AI credit. Unauthorized, insufficient-scope, and rate-limit failures are returned as MCP protocol errors by the host.

Prompt templates

PromptArgumentsPurpose
optimize-current-promptprompt; optional targetModelInstruct the agent to call optimize_prompt with one variation and return the best result
generate-ready-to-use-promptgoal; optional targetModelInstruct the agent to call generate_prompt with outputMode ready_to_use

Client support differs

A client may support tools while not exposing MCP prompts or resources in its interface. This does not mean the PrompTessor server failed to register them.

{
  "jsonrpc": "2.0",
  "id": 12,
  "method": "prompts/list",
  "params": {}
}

Prompt templates do not run tools by themselves

prompts/get returns messages that instruct the host agent to call a PrompTessor tool. The host still decides whether to call it, asks for any required approval, and enforces the tool OAuth scope.