@langwatch/mcp-server 2.1.0 → 2.2.0

package/dist/{delete-workflow-NG2VXYYM.js → delete-workflow-5J46T73T.js}

@@ -1,8 +1,8 @@
 import {
   deleteWorkflow
-} from "./chunk-UFOYFYI7.js";
-import "./chunk-GELJ2MQR.js";
-import "./chunk-WT4PO4A3.js";
+} from "./chunk-NKMTIHIJ.js";
+import "./chunk-UWLE5T2F.js";
+import "./chunk-H6WGMG2T.js";
 import "./chunk-SAS5OB7F.js";
 
 // src/tools/delete-workflow.ts
@@ -15,4 +15,4 @@ async function handleDeleteWorkflow(params) {
 export {
   handleDeleteWorkflow
 };
-//# sourceMappingURL=delete-workflow-NG2VXYYM.js.map
+//# sourceMappingURL=delete-workflow-5J46T73T.js.map

package/dist/delete-workflow-JSWABL5F.js

@@ -0,0 +1,17 @@
+import {
+  deleteWorkflow
+} from "./chunk-NKMTIHIJ.js";
+import "./chunk-UWLE5T2F.js";
+import "./chunk-H6WGMG2T.js";
+
+// src/tools/delete-workflow.ts
+async function handleDeleteWorkflow(params) {
+  const result = await deleteWorkflow(params.id);
+  return `Workflow archived successfully!
+
+**ID**: ${result.id}`;
+}
+export {
+  handleDeleteWorkflow
+};
+//# sourceMappingURL=delete-workflow-JSWABL5F.js.map

package/dist/discover-evaluator-schema-UHWA5HVL.js

@@ -0,0 +1,120 @@
+import "./chunk-SAS5OB7F.js";
+
+// src/tools/discover-evaluator-schema.ts
+import {
+  AVAILABLE_EVALUATORS
+} from "@langwatch/evaluator-contract";
+function formatEvaluatorSchema(evaluatorType) {
+  if (evaluatorType) {
+    return formatEvaluatorDetail(evaluatorType);
+  }
+  return formatEvaluatorOverview();
+}
+function formatEvaluatorOverview() {
+  const lines = [];
+  lines.push("# Available Evaluator Types\n");
+  const byCategory = /* @__PURE__ */ new Map();
+  for (const [type, def] of Object.entries(AVAILABLE_EVALUATORS)) {
+    const evalDef = def;
+    const oneLine = extractFirstLine(evalDef.description);
+    const entry = { type, name: evalDef.name, description: oneLine };
+    const list = byCategory.get(evalDef.category) ?? [];
+    list.push(entry);
+    byCategory.set(evalDef.category, list);
+  }
+  for (const [category, entries] of byCategory) {
+    lines.push(`## ${category}
+`);
+    for (const entry of entries) {
+      lines.push(`- **${entry.type}** (${entry.name}): ${entry.description}`);
+    }
+    lines.push("");
+  }
+  lines.push(
+    "> Use `discover_schema({ category: 'evaluators', evaluatorType: '<type>' })` for full details on a specific evaluator type."
+  );
+  return lines.join("\n");
+}
+function settingsSection(settingsEntries) {
+  if (settingsEntries.length === 0) return [];
+  const lines = ["\n## Settings\n"];
+  for (const [key, setting] of settingsEntries) {
+    const s = setting;
+    const defaultStr = JSON.stringify(s.default);
+    const desc = s.description ? ` - ${s.description}` : "";
+    lines.push(`- **${key}**${desc}`);
+    lines.push(`  Default: \`${defaultStr}\``);
+  }
+  return lines;
+}
+function formatEvaluatorDetail(evaluatorType) {
+  const def = AVAILABLE_EVALUATORS[evaluatorType];
+  if (!def) {
+    return `Unknown evaluator type: "${evaluatorType}". Use \`discover_schema({ category: 'evaluators' })\` to see all available types.`;
+  }
+  const lines = [];
+  lines.push(`# ${def.name} (\`${evaluatorType}\`)
+`);
+  lines.push(`**Category**: ${def.category}`);
+  lines.push(`**Is Guardrail**: ${def.isGuardrail ? "Yes" : "No"}`);
+  if (def.docsUrl) {
+    lines.push(`**Docs**: ${def.docsUrl}`);
+  }
+  lines.push("");
+  lines.push(`## Description
+`);
+  lines.push(def.description.trim());
+  lines.push("\n## Fields\n");
+  if (def.requiredFields.length > 0) {
+    lines.push(`**Required**: ${def.requiredFields.join(", ")}`);
+  } else {
+    lines.push("**Required**: none");
+  }
+  if (def.optionalFields.length > 0) {
+    lines.push(`**Optional**: ${def.optionalFields.join(", ")}`);
+  }
+  const settingsEntries = Object.entries(def.settings);
+  lines.push(...settingsSection(settingsEntries));
+  if (def.envVars.length > 0) {
+    lines.push("\n## Required Environment Variables\n");
+    for (const envVar of def.envVars) {
+      lines.push(`- \`${envVar}\``);
+    }
+  }
+  const resultEntries = Object.entries(def.result);
+  if (resultEntries.length > 0) {
+    lines.push("\n## Result Fields\n");
+    for (const [key, value] of resultEntries) {
+      const v = value;
+      lines.push(`- **${key}**: ${v.description}`);
+    }
+  }
+  lines.push("\n## Usage Example\n");
+  lines.push("```json");
+  lines.push(
+    JSON.stringify(
+      {
+        evaluatorType,
+        settings: Object.fromEntries(
+          settingsEntries.map(([key, setting]) => [key, setting.default])
+        )
+      },
+      null,
+      2
+    )
+  );
+  lines.push("```");
+  return lines.join("\n");
+}
+function extractFirstLine(description) {
+  const trimmed = description.trim();
+  const firstLine = trimmed.split("\n")[0]?.trim() ?? trimmed;
+  if (firstLine.length > 120) {
+    return firstLine.slice(0, 117) + "...";
+  }
+  return firstLine;
+}
+export {
+  formatEvaluatorSchema
+};
+//# sourceMappingURL=discover-evaluator-schema-UHWA5HVL.js.map

package/dist/discover-evaluator-schema-XKCLLCB3.js

@@ -0,0 +1,118 @@
+// src/tools/discover-evaluator-schema.ts
+import {
+  AVAILABLE_EVALUATORS
+} from "@langwatch/evaluator-contract";
+function formatEvaluatorSchema(evaluatorType) {
+  if (evaluatorType) {
+    return formatEvaluatorDetail(evaluatorType);
+  }
+  return formatEvaluatorOverview();
+}
+function formatEvaluatorOverview() {
+  const lines = [];
+  lines.push("# Available Evaluator Types\n");
+  const byCategory = /* @__PURE__ */ new Map();
+  for (const [type, def] of Object.entries(AVAILABLE_EVALUATORS)) {
+    const evalDef = def;
+    const oneLine = extractFirstLine(evalDef.description);
+    const entry = { type, name: evalDef.name, description: oneLine };
+    const list = byCategory.get(evalDef.category) ?? [];
+    list.push(entry);
+    byCategory.set(evalDef.category, list);
+  }
+  for (const [category, entries] of byCategory) {
+    lines.push(`## ${category}
+`);
+    for (const entry of entries) {
+      lines.push(`- **${entry.type}** (${entry.name}): ${entry.description}`);
+    }
+    lines.push("");
+  }
+  lines.push(
+    "> Use `discover_schema({ category: 'evaluators', evaluatorType: '<type>' })` for full details on a specific evaluator type."
+  );
+  return lines.join("\n");
+}
+function settingsSection(settingsEntries) {
+  if (settingsEntries.length === 0) return [];
+  const lines = ["\n## Settings\n"];
+  for (const [key, setting] of settingsEntries) {
+    const s = setting;
+    const defaultStr = JSON.stringify(s.default);
+    const desc = s.description ? ` - ${s.description}` : "";
+    lines.push(`- **${key}**${desc}`);
+    lines.push(`  Default: \`${defaultStr}\``);
+  }
+  return lines;
+}
+function formatEvaluatorDetail(evaluatorType) {
+  const def = AVAILABLE_EVALUATORS[evaluatorType];
+  if (!def) {
+    return `Unknown evaluator type: "${evaluatorType}". Use \`discover_schema({ category: 'evaluators' })\` to see all available types.`;
+  }
+  const lines = [];
+  lines.push(`# ${def.name} (\`${evaluatorType}\`)
+`);
+  lines.push(`**Category**: ${def.category}`);
+  lines.push(`**Is Guardrail**: ${def.isGuardrail ? "Yes" : "No"}`);
+  if (def.docsUrl) {
+    lines.push(`**Docs**: ${def.docsUrl}`);
+  }
+  lines.push("");
+  lines.push(`## Description
+`);
+  lines.push(def.description.trim());
+  lines.push("\n## Fields\n");
+  if (def.requiredFields.length > 0) {
+    lines.push(`**Required**: ${def.requiredFields.join(", ")}`);
+  } else {
+    lines.push("**Required**: none");
+  }
+  if (def.optionalFields.length > 0) {
+    lines.push(`**Optional**: ${def.optionalFields.join(", ")}`);
+  }
+  const settingsEntries = Object.entries(def.settings);
+  lines.push(...settingsSection(settingsEntries));
+  if (def.envVars.length > 0) {
+    lines.push("\n## Required Environment Variables\n");
+    for (const envVar of def.envVars) {
+      lines.push(`- \`${envVar}\``);
+    }
+  }
+  const resultEntries = Object.entries(def.result);
+  if (resultEntries.length > 0) {
+    lines.push("\n## Result Fields\n");
+    for (const [key, value] of resultEntries) {
+      const v = value;
+      lines.push(`- **${key}**: ${v.description}`);
+    }
+  }
+  lines.push("\n## Usage Example\n");
+  lines.push("```json");
+  lines.push(
+    JSON.stringify(
+      {
+        evaluatorType,
+        settings: Object.fromEntries(
+          settingsEntries.map(([key, setting]) => [key, setting.default])
+        )
+      },
+      null,
+      2
+    )
+  );
+  lines.push("```");
+  return lines.join("\n");
+}
+function extractFirstLine(description) {
+  const trimmed = description.trim();
+  const firstLine = trimmed.split("\n")[0]?.trim() ?? trimmed;
+  if (firstLine.length > 120) {
+    return firstLine.slice(0, 117) + "...";
+  }
+  return firstLine;
+}
+export {
+  formatEvaluatorSchema
+};
+//# sourceMappingURL=discover-evaluator-schema-XKCLLCB3.js.map

package/dist/discover-scenario-schema-NOJY6S7E.js

@@ -0,0 +1,137 @@
+// src/tools/discover-scenario-schema.ts
+function formatScenarioSchema() {
+  const lines = [];
+  lines.push("# Scenario Schema\n");
+  lines.push("## Fields\n");
+  lines.push(
+    '- **name** (required): A short, descriptive name (e.g., "billing dispute resolution", "password reset with 2FA unavailable")'
+  );
+  lines.push(
+    "- **situation** (required): The context that guides the user simulator \u2014 who the user is, what they want, and any constraints (see Writing a Good Situation below)"
+  );
+  lines.push(
+    "- **criteria** (array of strings): Pass/fail conditions a judge evaluates the agent against (see Writing Good Criteria below)"
+  );
+  lines.push(
+    '- **labels** (array of strings): Tags for organizing scenarios (e.g., "auth", "happy-path", "edge-case")'
+  );
+  lines.push(
+    "- **testSuiteId** (string): The test suite the scenario is filed in. Every project has a Default suite."
+  );
+  lines.push(
+    `- **fields** (object): The scenario's value for each field its test suite declares, by identifier (e.g. { "golden_sql": "SELECT ..." }). A text field takes a string, a number field a number, a boolean field true or false. A field the suite does not declare is refused. A blank value skips the evaluators that read it.`
+  );
+  lines.push("\n## Test Suite Fields\n");
+  lines.push(
+    "A test suite reads like a dataset: it declares typed fields beyond the situation and the criteria, and every scenario filed in it carries one value per field. Declare them with `platform_create_test_suite` or `platform_update_test_suite`:"
+  );
+  lines.push("```json");
+  lines.push(
+    '{ "fields": [{ "identifier": "golden_sql", "type": "text" }, { "identifier": "row_limit", "type": "number" }] }'
+  );
+  lines.push("```");
+  lines.push(
+    "- **identifier**: lowercase letters, digits and underscores, starting with a letter. situation, criteria, name, input and output are reserved."
+  );
+  lines.push("- **type**: text, number or boolean.");
+  lines.push(
+    "- On update, the list replaces the one the suite holds. A field an attached evaluator still reads cannot be removed."
+  );
+  lines.push("\n## Evaluator Attachments\n");
+  lines.push(
+    "A saved evaluator attaches to a test suite (and, for checks that read only the conversation or the trace, to a run plan). After every scenario run the platform runs it and stores one result per evaluator on the run. A **required** pass/fail evaluator that fails fails the scenario; a score-only evaluator reports beside the verdict and never gates."
+  );
+  lines.push("```json");
+  lines.push('{ "evaluators": [{');
+  lines.push('  "evaluatorId": "<id from platform_list_evaluators>",');
+  lines.push('  "required": true,');
+  lines.push('  "mappings": {');
+  lines.push(
+    '    "output": { "type": "source", "sourceId": "trace", "path": ["tool_calls", "run_sql", "input"] },'
+  );
+  lines.push(
+    '    "expected_output": { "type": "source", "sourceId": "scenario", "path": ["fields", "golden_sql"] }'
+  );
+  lines.push("  }");
+  lines.push("}] }");
+  lines.push("```");
+  lines.push(
+    "Each mapping names where one evaluator input reads its value. Read the evaluator's inputs from `platform_get_evaluator`; every required input needs a mapping before a run. The paths:"
+  );
+  lines.push(
+    '- **conversation**: `["first_user_message"]`, `["last_agent_message"]`, `["transcript"]` (role: content lines) or `["messages"]` (JSON)'
+  );
+  lines.push(
+    '- **scenario**: `["situation"]`, `["criteria"]` or `["fields", "<identifier>"]` (a field the suite declares; not on a run plan)'
+  );
+  lines.push(
+    '- **trace**: `["contexts"]` (the retrieved contexts of the run\'s traces) or `["tool_calls", "<toolName>", "input" | "output"]` (the last matching tool call)'
+  );
+  lines.push('- a literal: `{ "type": "value", "value": "..." }`');
+  lines.push(
+    "A scenario whose field is blank skips the evaluators that read it, with a reason. A trace with no such tool call or no contexts fails them, with a reason. The results come back under `results.evaluations` on `platform_get_simulation_run`."
+  );
+  lines.push("\n## Writing a Good Situation\n");
+  lines.push("The situation drives the user simulator. Include these elements:");
+  lines.push(
+    "- **Persona**: Who is the user? (e.g., a stressed small business owner, a confused teenager)"
+  );
+  lines.push("- **Emotional state**: How are they feeling? (e.g., frustrated, anxious, impatient)");
+  lines.push("- **Background/Context**: What happened before this conversation?");
+  lines.push("- **Intent**: What do they want to accomplish?");
+  lines.push(
+    "- **Constraints**: What limitations do they have? (e.g., no phone for 2FA, unfamiliar with technical terms)"
+  );
+  lines.push("\nExample:");
+  lines.push("```");
+  lines.push("User is a small business owner stressed about tax deadline.");
+  lines.push("They need help categorizing expenses but aren't familiar with");
+  lines.push("accounting terms. They appreciate patient explanations and examples.");
+  lines.push("They have a spreadsheet of transactions but aren't sure which");
+  lines.push("categories apply to their consulting business.");
+  lines.push("```");
+  lines.push("\n## Writing Good Criteria\n");
+  lines.push(
+    "Criteria are what the judge uses to pass or fail the agent. Each criterion should be:"
+  );
+  lines.push('- **Specific and testable** \u2014 not vague like "responds helpfully"');
+  lines.push(
+    "- **Behavioral** \u2014 describes what the agent should *do*, not how it works internally"
+  );
+  lines.push("- **Independent** \u2014 each criterion checks one thing");
+  lines.push("\nGood criteria patterns:");
+  lines.push(
+    `- **Information gathering**: "Agent asks for the user's account number before proceeding"`
+  );
+  lines.push(
+    '- **Safety/guardrails**: "Agent does not reveal internal system details or error stack traces"'
+  );
+  lines.push(
+    '- **Clarification**: "Agent asks clarifying questions before taking irreversible action"'
+  );
+  lines.push('- **Tone**: "Agent maintains a professional and empathetic tone throughout"');
+  lines.push(
+    '- **Completeness**: "Agent confirms the user understands the solution before ending"'
+  );
+  lines.push(
+    '- **Domain-specific**: "Agent recommends releasing a wild frog rather than keeping it as a pet"'
+  );
+  lines.push("\nAvoid vague criteria like:");
+  lines.push('- "Responds correctly" \u2014 correct how?');
+  lines.push('- "Is helpful" \u2014 helpful in what way?');
+  lines.push('- "Works well" \u2014 not testable');
+  lines.push("\n## Target Types\n");
+  lines.push("Scenarios can target different execution backends:");
+  lines.push("- **prompt**: Test a prompt template with variable substitution");
+  lines.push("- **http**: Test an HTTP endpoint (e.g., a deployed agent API)");
+  lines.push("- **code**: Test a code function directly");
+  lines.push("\n## Tips\n");
+  lines.push("- Start simple, then layer complexity (add constraints, edge cases)");
+  lines.push("- Test edge cases: user changes their mind, gives ambiguous input, makes mistakes");
+  lines.push("- Use `fetch_scenario_docs` for the full authoring guide and advanced patterns");
+  return lines.join("\n");
+}
+export {
+  formatScenarioSchema
+};
+//# sourceMappingURL=discover-scenario-schema-NOJY6S7E.js.map

package/dist/{discover-scenario-schema-W7KBXA4J.js → discover-scenario-schema-OOFIQKOG.js}

@@ -28,7 +28,9 @@ function formatScenarioSchema() {
     "A test suite reads like a dataset: it declares typed fields beyond the situation and the criteria, and every scenario filed in it carries one value per field. Declare them with `platform_create_test_suite` or `platform_update_test_suite`:"
   );
   lines.push("```json");
-  lines.push('{ "fields": [{ "identifier": "golden_sql", "type": "text" }, { "identifier": "row_limit", "type": "number" }] }');
+  lines.push(
+    '{ "fields": [{ "identifier": "golden_sql", "type": "text" }, { "identifier": "row_limit", "type": "number" }] }'
+  );
   lines.push("```");
   lines.push(
     "- **identifier**: lowercase letters, digits and underscores, starting with a letter. situation, criteria, name, input and output are reserved."
@@ -46,8 +48,12 @@ function formatScenarioSchema() {
   lines.push('  "evaluatorId": "<id from platform_list_evaluators>",');
   lines.push('  "required": true,');
   lines.push('  "mappings": {');
-  lines.push('    "output": { "type": "source", "sourceId": "trace", "path": ["tool_calls", "run_sql", "input"] },');
-  lines.push('    "expected_output": { "type": "source", "sourceId": "scenario", "path": ["fields", "golden_sql"] }');
+  lines.push(
+    '    "output": { "type": "source", "sourceId": "trace", "path": ["tool_calls", "run_sql", "input"] },'
+  );
+  lines.push(
+    '    "expected_output": { "type": "source", "sourceId": "scenario", "path": ["fields", "golden_sql"] }'
+  );
   lines.push("  }");
   lines.push("}] }");
   lines.push("```");
@@ -63,21 +69,21 @@ function formatScenarioSchema() {
   lines.push(
     '- **trace**: `["contexts"]` (the retrieved contexts of the run\'s traces) or `["tool_calls", "<toolName>", "input" | "output"]` (the last matching tool call)'
   );
-  lines.push(
-    '- a literal: `{ "type": "value", "value": "..." }`'
-  );
+  lines.push('- a literal: `{ "type": "value", "value": "..." }`');
   lines.push(
     "A scenario whose field is blank skips the evaluators that read it, with a reason. A trace with no such tool call or no contexts fails them, with a reason. The results come back under `results.evaluations` on `platform_get_simulation_run`."
   );
   lines.push("\n## Writing a Good Situation\n");
+  lines.push("The situation drives the user simulator. Include these elements:");
   lines.push(
-    "The situation drives the user simulator. Include these elements:"
+    "- **Persona**: Who is the user? (e.g., a stressed small business owner, a confused teenager)"
   );
-  lines.push("- **Persona**: Who is the user? (e.g., a stressed small business owner, a confused teenager)");
   lines.push("- **Emotional state**: How are they feeling? (e.g., frustrated, anxious, impatient)");
   lines.push("- **Background/Context**: What happened before this conversation?");
   lines.push("- **Intent**: What do they want to accomplish?");
-  lines.push("- **Constraints**: What limitations do they have? (e.g., no phone for 2FA, unfamiliar with technical terms)");
+  lines.push(
+    "- **Constraints**: What limitations do they have? (e.g., no phone for 2FA, unfamiliar with technical terms)"
+  );
   lines.push("\nExample:");
   lines.push("```");
   lines.push("User is a small business owner stressed about tax deadline.");
@@ -87,17 +93,31 @@ function formatScenarioSchema() {
   lines.push("categories apply to their consulting business.");
   lines.push("```");
   lines.push("\n## Writing Good Criteria\n");
-  lines.push("Criteria are what the judge uses to pass or fail the agent. Each criterion should be:");
+  lines.push(
+    "Criteria are what the judge uses to pass or fail the agent. Each criterion should be:"
+  );
   lines.push('- **Specific and testable** \u2014 not vague like "responds helpfully"');
-  lines.push("- **Behavioral** \u2014 describes what the agent should *do*, not how it works internally");
+  lines.push(
+    "- **Behavioral** \u2014 describes what the agent should *do*, not how it works internally"
+  );
   lines.push("- **Independent** \u2014 each criterion checks one thing");
   lines.push("\nGood criteria patterns:");
-  lines.push(`- **Information gathering**: "Agent asks for the user's account number before proceeding"`);
-  lines.push('- **Safety/guardrails**: "Agent does not reveal internal system details or error stack traces"');
-  lines.push('- **Clarification**: "Agent asks clarifying questions before taking irreversible action"');
+  lines.push(
+    `- **Information gathering**: "Agent asks for the user's account number before proceeding"`
+  );
+  lines.push(
+    '- **Safety/guardrails**: "Agent does not reveal internal system details or error stack traces"'
+  );
+  lines.push(
+    '- **Clarification**: "Agent asks clarifying questions before taking irreversible action"'
+  );
   lines.push('- **Tone**: "Agent maintains a professional and empathetic tone throughout"');
-  lines.push('- **Completeness**: "Agent confirms the user understands the solution before ending"');
-  lines.push('- **Domain-specific**: "Agent recommends releasing a wild frog rather than keeping it as a pet"');
+  lines.push(
+    '- **Completeness**: "Agent confirms the user understands the solution before ending"'
+  );
+  lines.push(
+    '- **Domain-specific**: "Agent recommends releasing a wild frog rather than keeping it as a pet"'
+  );
   lines.push("\nAvoid vague criteria like:");
   lines.push('- "Responds correctly" \u2014 correct how?');
   lines.push('- "Is helpful" \u2014 helpful in what way?');
@@ -116,4 +136,4 @@ function formatScenarioSchema() {
 export {
   formatScenarioSchema
 };
-//# sourceMappingURL=discover-scenario-schema-W7KBXA4J.js.map
+//# sourceMappingURL=discover-scenario-schema-OOFIQKOG.js.map