MCP for Claude & ChatGPT

Manage tests

tuqui_evals

Destructive

The tests of an agent or a project: cases that run through the real agent and say, case by case, whether it still answers as expected. Cases are written as YAML (see 'yaml'). Name what is tested with 'agent_id' or 'project_id' (exactly one). action='list' returns its cases; 'get' one by 'case_id', with its YAML; 'export' the whole set as one YAML text; 'save' loads 'yaml' into the set (the name identifies a case: a known name replaces it, keeping its runs; up to 50 per agent or project and 500000 characters of text) or, with 'case_id', rewrites that case from 'yaml'; 'delete' removes 'case_id'. 'run' starts a run of the active cases in the background, or of only 'case_id' (writes are held back and recorded, never executed); 'runs' lists the latest runs and the conditions each ran under; 'report' reads one run case by case ('run_id', or the latest), with which checks failed and how it did in the previous run under the same conditions, and what each case answered only for whoever launched the run; 'cancel' stops 'run_id'; 'history' lists every run of 'case_id'. Only someone who can edit the agent or the project manages its tests, and the agent never sees them.

Tool call template

Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {
      "action": "<action>"
    },
    "name": "tuqui_evals"
  }
}

Input Schema

Name	Required	Description
`yaml`	No	The case(s) as YAML, one per document, several separated by '---'. A case: 'name'; 'input' (the message) or 'trigger' (a record that fires it, as the webhook does: "res.partner #43" quoted, since unquoted YAML reads #43 as a comment, or {model, id}); 'expect' with any of: contains / not_contains (texts the reply must or must not contain; a value between slashes, /like this/, is a regex), equals, tools / no_tools (tools it must or must not call; an entry may be {tool, args} to require those arguments), writes (a list of {tool, args}: writes it must try, with those arguments as a subset; an argument may be '/regex/' or {$ne: x} instead of a fixed value; writes are never executed in a test, they are recorded), judge (a yes/no question a judgment model answers about the reply, or {criterion, threshold} to pass from another probability than 0.7; prefer the exact checks, the judge is for free text); optional 'active: false'. Comments are kept. Example: name: Teléfono de Acme input: ¿Cuál es el teléfono de Acme? expect: contains: ['996-3829'] tools: [odoo_search_read]
`action`	Yes	REQUIRED. What to do with the tests.
`run_id`	No	report / cancel: the run. report without it reads the latest.
`case_id`	No	get / delete / history; save to rewrite one case; run to run only that case.
`agent_id`	No	The agent whose tests these are.
`project_id`	No	The project whose tests these are.

Schema Changelog

Changes observed during successful MCP inspections.

AddedNov 6, 2026

TDQS

A4.4/5.0

Behavior3/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations already flag destructiveHint=true and openWorldHint=false, so the safety profile is partially given. The description adds real non-obvious behavior beyond that: writes during a run are 'held back and recorded, never executed', save replaces a known case name while keeping its runs, and there are hard limits of 50 cases and 500000 characters. This is substantial added context, but return shapes and failure behavior are not described.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

Purpose is front-loaded and nearly every clause carries information, with no filler sentences. The cost is that the action map is delivered as one dense semicolon-chained paragraph, which makes scanning for a specific action harder than a structured list would be.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness4/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

For a 10-action, 6-parameter tool with no output schema, the description covers every action and states what each returns at a high level plus permission requirements. It stops short of describing the shape of 'list'/'runs'/'history' output, which would matter since no output schema exists.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters5/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Schema coverage is 100%, so a baseline of 3 would apply, but the description goes well beyond it by mapping each parameter to the actions that consume it ('run_id' for report/cancel, 'case_id' for get/delete/history, save-rewrite, run-only-that-case). It also explains the format of 'yaml' with an inline example and quoting caveats that the schema alone would not convey.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The first sentence names the specific resource ('the tests of an agent or a project') and what it does ('cases that run through the real agent and say, case by case, whether it still answers as expected'). It is clearly distinct from siblings like tuqui_agent or judge, and the subsequent action map removes any ambiguity about scope.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

Every action is given a when-to-use definition ('list' returns its cases, 'get' one by case_id with its YAML, 'report' reads one run case by case), including selection rules such as 'exactly one' of agent_id or project_id and the note to prefer exact checks over the judge. Prerequisites are explicit: only someone who can edit the agent or project can manage tests, and the agent never sees them.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

All MCP for Claude & ChatGPT tools Inspect

Try in Browser

One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

Glama MCP Gateway

Add one secure layer between your agents and this server.