@langwatch/mcp-server 0.7.0 → 0.9.0

package/src/tools/get-experiment-results.ts

@@ -0,0 +1,273 @@
+import { LangWatchApiError, makeRequest } from "../langwatch-api.js";
+
+interface DatasetEntry {
+  index: number;
+  targetId?: string | null;
+  entry: Record<string, unknown>;
+  predicted?: Record<string, unknown>;
+  cost?: number | null;
+  duration?: number | null;
+  error?: string | null;
+  traceId?: string | null;
+}
+
+interface EvaluationItem {
+  evaluator: string;
+  name?: string | null;
+  index: number;
+  targetId?: string | null;
+  status: "processed" | "skipped" | "error";
+  score?: number | null;
+  label?: string | null;
+  passed?: boolean | null;
+  details?: string | null;
+  inputs?: Record<string, unknown> | null;
+}
+
+interface EvaluationRunResults {
+  experimentId: string;
+  runId: string;
+  projectId: string;
+  progress?: number | null;
+  total?: number | null;
+  dataset: DatasetEntry[];
+  evaluations: EvaluationItem[];
+  timestamps: {
+    createdAt: number;
+    updatedAt: number;
+    finishedAt?: number | null;
+    stoppedAt?: number | null;
+  };
+}
+
+const DEFAULT_ROW_CAP = 50;
+
+const rowKey = (index: number, targetId?: string | null): string =>
+  `${index}:${targetId ?? ""}`;
+
+const summarizeEntry = (entry: Record<string, unknown>): string => {
+  const candidates = ["input", "question", "query", "prompt", "user"];
+  for (const key of candidates) {
+    const value = entry[key];
+    if (typeof value === "string" && value.length > 0) {
+      return value.length > 80 ? `${value.slice(0, 77)}...` : value;
+    }
+  }
+  return "";
+};
+
+const isFailedEvaluation = (e: EvaluationItem): boolean =>
+  e.status === "error" || e.passed === false;
+
+const isFailedRow = ({
+  entry,
+  evaluations,
+}: {
+  entry: DatasetEntry;
+  evaluations: EvaluationItem[];
+}): boolean => Boolean(entry.error) || evaluations.some(isFailedEvaluation);
+
+export async function handleExperimentResults(params: {
+  runId: string;
+  experimentSlug?: string;
+  filter?: "all" | "failed";
+  evaluator?: string;
+  limit?: number;
+}): Promise<string> {
+  const filter = params.filter ?? "all";
+  const evaluatorFilter = params.evaluator?.trim();
+  const limit =
+    typeof params.limit === "number" && params.limit > 0
+      ? Math.min(params.limit, DEFAULT_ROW_CAP)
+      : DEFAULT_ROW_CAP;
+
+  const search = new URLSearchParams();
+  if (params.experimentSlug) search.set("experimentSlug", params.experimentSlug);
+  const qs = search.toString() ? `?${search.toString()}` : "";
+
+  let results: EvaluationRunResults | null;
+  try {
+    results = (await makeRequest(
+      "GET",
+      `/api/experiments/runs/${encodeURIComponent(params.runId)}/results${qs}`,
+    )) as EvaluationRunResults;
+  } catch (error) {
+    const message = error instanceof Error ? error.message : String(error);
+    const status =
+      error instanceof LangWatchApiError
+        ? error.status
+        : error && typeof error === "object" &&
+            "status" in error &&
+            typeof (error as { status?: unknown }).status === "number"
+          ? (error as { status: number }).status
+          : undefined;
+    if (
+      status === 404 ||
+      (status === undefined && /404|not found/i.test(message))
+    ) {
+      return [
+        `# Evaluation Results: ${params.runId}`,
+        "",
+        "**Status**: not found",
+        "",
+        `Could not load results for run \`${params.runId}\`. The run may still be in progress, or the id may be incorrect.`,
+        "",
+        "> Try \`platform_experiment_status\` first to confirm the run id and that it is completed.",
+      ].join("\n");
+    }
+    throw error;
+  }
+
+  if (!results || (results.timestamps.finishedAt == null && results.timestamps.stoppedAt == null)) {
+    return [
+      `# Evaluation Results: ${params.runId}`,
+      "",
+      "**Status**: results are not yet available",
+      "",
+      `Results for run \`${params.runId}\` are not ready yet.`,
+      "",
+      "> Try `platform_experiment_status` first to confirm the run id and that it is completed.",
+    ].join("\n");
+  }
+
+  // Group evaluations by target-scoped row key, applying evaluator filter.
+  const evaluationsByRow = new Map<string, EvaluationItem[]>();
+  for (const evaluation of results.evaluations) {
+    if (evaluatorFilter && evaluation.evaluator !== evaluatorFilter) continue;
+    const key = rowKey(evaluation.index, evaluation.targetId);
+    const list = evaluationsByRow.get(key) ?? [];
+    list.push(evaluation);
+    evaluationsByRow.set(key, list);
+  }
+
+  const evaluatorNames = Array.from(
+    new Set(
+      (evaluatorFilter
+        ? results.evaluations.filter((e) => e.evaluator === evaluatorFilter)
+        : results.evaluations
+      ).map((e) => e.evaluator),
+    ),
+  );
+
+  let rows = results.dataset.map((entry) => ({
+    entry,
+    evaluations: evaluationsByRow.get(rowKey(entry.index, entry.targetId)) ?? [],
+  }));
+
+  if (filter === "failed") {
+    rows = rows.filter((r) =>
+      isFailedRow({ entry: r.entry, evaluations: r.evaluations }),
+    );
+  }
+
+  const totalMatching = rows.length;
+  const rowsForSummary = rows;
+  const truncated = rows.length > limit;
+  rows = rows.slice(0, limit);
+
+  // Per-evaluator stats across the filtered rows (before truncation),
+  // so the summary matches the displayed subset when filter="failed".
+  const evaluatorAverages = new Map<
+    string,
+    { sum: number; count: number; passed: number; failed: number; errored: number }
+  >();
+  for (const r of rowsForSummary) {
+    for (const e of r.evaluations) {
+      const stats =
+        evaluatorAverages.get(e.evaluator) ?? {
+          sum: 0,
+          count: 0,
+          passed: 0,
+          failed: 0,
+          errored: 0,
+        };
+      if (typeof e.score === "number") {
+        stats.sum += e.score;
+        stats.count += 1;
+      }
+      if (e.status === "error") stats.errored += 1;
+      else if (e.passed === true) stats.passed += 1;
+      else if (e.passed === false) stats.failed += 1;
+      evaluatorAverages.set(e.evaluator, stats);
+    }
+  }
+
+  const lines: string[] = [];
+  lines.push(`# Evaluation Results: ${results.runId}`);
+  lines.push("");
+  lines.push(`**Experiment**: ${results.experimentId}`);
+  lines.push(`**Total rows**: ${results.dataset.length}`);
+  lines.push(`**Total evaluations**: ${results.evaluations.length}`);
+  if (filter === "failed") {
+    lines.push(`**Filter**: failed only`);
+  }
+  if (evaluatorFilter) {
+    lines.push(`**Evaluator filter**: ${evaluatorFilter}`);
+  }
+  lines.push("");
+
+  if (evaluatorAverages.size > 0) {
+    lines.push("## Evaluator Summary");
+    lines.push("");
+    lines.push("| Evaluator | Avg Score | Passed | Failed | Errored |");
+    lines.push("| --- | --- | --- | --- | --- |");
+    for (const [name, stats] of evaluatorAverages) {
+      const avg =
+        stats.count > 0 ? (stats.sum / stats.count).toFixed(3) : "—";
+      lines.push(
+        `| ${name} | ${avg} | ${stats.passed} | ${stats.failed} | ${stats.errored} |`,
+      );
+    }
+    lines.push("");
+  }
+
+  if (rows.length === 0) {
+    lines.push("_No rows matched the filter._");
+    return lines.join("\n");
+  }
+
+  lines.push(
+    `## Rows (${rows.length}${truncated ? ` of ${totalMatching}` : ""})`,
+  );
+  lines.push("");
+
+  for (const { entry, evaluations } of rows) {
+    const summary = summarizeEntry(entry.entry);
+    lines.push(`### Row #${entry.index}${summary ? ` — ${summary}` : ""}`);
+    if (entry.error) {
+      lines.push(`- **Error**: ${entry.error}`);
+    }
+    if (entry.traceId) {
+      lines.push(`- **Trace ID**: \`${entry.traceId}\``);
+    }
+    if (evaluations.length === 0) {
+      lines.push(
+        evaluatorNames.length === 0
+          ? "- _No evaluations recorded_"
+          : "- _No evaluations recorded for this row_",
+      );
+    }
+    for (const e of evaluations) {
+      const parts: string[] = [`**${e.evaluator}**`];
+      if (e.status === "error") parts.push("status=error");
+      else if (typeof e.score === "number") parts.push(`score=${e.score.toFixed(3)}`);
+      if (typeof e.passed === "boolean") {
+        parts.push(`passed=${e.passed ? "yes" : "no"}`);
+      }
+      if (e.label) parts.push(`label=${e.label}`);
+      lines.push(`- ${parts.join(" · ")}`);
+      if (e.details) {
+        lines.push(`  - details: ${e.details}`);
+      }
+    }
+    lines.push("");
+  }
+
+  if (truncated) {
+    lines.push(
+      `> Output truncated to ${limit} rows of ${totalMatching} matching to protect the agent's context window. Pass \`limit\` to expand or filter further.`,
+    );
+  }
+
+  return lines.join("\n");
+}

package/src/tools/get-project.ts

@@ -0,0 +1,20 @@
+import { getProject as apiGetProject } from "../langwatch-api-projects.js";
+
+export async function handleGetProject(params: {
+  id: string;
+}): Promise<string> {
+  const project = await apiGetProject(params.id);
+
+  const lines: string[] = [];
+  lines.push(`# ${project.name}\n`);
+  lines.push(`**ID**: ${project.id}`);
+  lines.push(`**Slug**: ${project.slug}`);
+  lines.push(`**Language**: ${project.language}`);
+  lines.push(`**Framework**: ${project.framework}`);
+  lines.push(`**Team ID**: ${project.teamId}`);
+  lines.push(`**PII Redaction**: ${project.piiRedactionLevel}`);
+  lines.push(`**Created**: ${project.createdAt}`);
+  lines.push(`**Updated**: ${project.updatedAt}`);
+
+  return lines.join("\n");
+}

package/src/tools/list-api-keys.ts

@@ -0,0 +1,33 @@
+import { listApiKeys as apiListApiKeys } from "../langwatch-api-api-keys.js";
+
+export async function handleListApiKeys(): Promise<string> {
+  const result = await apiListApiKeys();
+  const keys = result.data;
+
+  if (!Array.isArray(keys) || keys.length === 0) {
+    return "No API keys found.\n\n> Tip: Use `platform_create_api_key` to create a new key.";
+  }
+
+  const lines: string[] = [];
+  lines.push(`# API Keys (${keys.length} total)\n`);
+
+  for (const k of keys) {
+    const status = k.revokedAt ? "REVOKED" : k.expiresAt && new Date(k.expiresAt) < new Date() ? "EXPIRED" : "ACTIVE";
+    lines.push(`## ${k.name}`);
+    lines.push(`**ID**: ${k.id}`);
+    lines.push(`**Status**: ${status}`);
+    if (k.description) lines.push(`**Description**: ${k.description}`);
+    if (k.expiresAt) lines.push(`**Expires**: ${k.expiresAt}`);
+    if (k.lastUsedAt) lines.push(`**Last Used**: ${k.lastUsedAt}`);
+    lines.push(`**Created**: ${k.createdAt}`);
+    if (k.roleBindings.length > 0) {
+      const bindings = k.roleBindings
+        .map((rb) => `${rb.role} on ${rb.scopeType}:${rb.scopeId}`)
+        .join(", ");
+      lines.push(`**Bindings**: ${bindings}`);
+    }
+    lines.push("");
+  }
+
+  return lines.join("\n");
+}

package/src/tools/list-experiment-runs.ts

@@ -0,0 +1,142 @@
+import { makeRequest } from "../langwatch-api.js";
+
+interface RunSummaryEvaluator {
+  name: string;
+  averageScore: number | null;
+  averagePassed?: number;
+}
+
+interface RunSummary {
+  evaluations: Record<string, RunSummaryEvaluator>;
+}
+
+interface ExperimentRunSummaryEntry {
+  experimentId: string;
+  runId: string;
+  workflowVersion: { id: string; version: string } | null;
+  timestamps: {
+    createdAt: number;
+    updatedAt: number;
+    finishedAt?: number | null;
+    stoppedAt?: number | null;
+  };
+  progress?: number | null;
+  total?: number | null;
+  summary: RunSummary;
+}
+
+interface ExperimentRunsListResponse {
+  experimentId: string;
+  experimentSlug: string;
+  runs: ExperimentRunSummaryEntry[];
+  pagination: {
+    page: number;
+    pageSize: number;
+    totalHits: number;
+    hasMore: boolean;
+  };
+}
+
+const DEFAULT_LIMIT = 25;
+const MAX_LIMIT = 100;
+
+const formatEpoch = (epoch: number | null | undefined): string => {
+  if (!epoch || !Number.isFinite(epoch)) return "—";
+  return new Date(epoch).toISOString().replace("T", " ").slice(0, 19) + " UTC";
+};
+
+const passRate = (evaluations: RunSummary["evaluations"]): string => {
+  const entries = Object.values(evaluations ?? {});
+  if (entries.length === 0) return "—";
+  const passed = entries.filter((e) => typeof e.averagePassed === "number");
+  if (passed.length === 0) {
+    const scored = entries.filter((e) => typeof e.averageScore === "number");
+    if (scored.length === 0) return "—";
+    const avg =
+      scored.reduce((sum, e) => sum + (e.averageScore ?? 0), 0) / scored.length;
+    return `${avg.toFixed(2)} avg`;
+  }
+  const avg =
+    passed.reduce((sum, e) => sum + (e.averagePassed ?? 0), 0) / passed.length;
+  return `${(avg * 100).toFixed(0)}% pass`;
+};
+
+const runStatus = (run: ExperimentRunSummaryEntry): string => {
+  if (run.timestamps.stoppedAt) return "stopped";
+  if (run.timestamps.finishedAt) return "completed";
+  return "running";
+};
+
+export async function handleExperimentListRuns(params: {
+  experimentSlug: string;
+  limit?: number;
+}): Promise<string> {
+  const requested =
+    typeof params.limit === "number" && params.limit > 0
+      ? params.limit
+      : DEFAULT_LIMIT;
+  const effectiveLimit = Math.min(requested, MAX_LIMIT);
+
+  const search = new URLSearchParams();
+  search.set("experimentSlug", params.experimentSlug);
+  search.set("pageSize", String(effectiveLimit));
+
+  let result: ExperimentRunsListResponse;
+  try {
+    result = (await makeRequest(
+      "GET",
+      `/api/experiments/runs?${search.toString()}`,
+    )) as ExperimentRunsListResponse;
+  } catch (error) {
+    const message = error instanceof Error ? error.message : String(error);
+    if (/404|not found/i.test(message)) {
+      return [
+        `# Evaluation Runs: ${params.experimentSlug}`,
+        "",
+        "**Status**: experiment not found",
+        "",
+        `No experiment with slug \`${params.experimentSlug}\` exists in this project.`,
+        "",
+        "> Run `platform_experiment_list` to discover available experiment slugs.",
+      ].join("\n");
+    }
+    throw error;
+  }
+
+  if (result.runs.length === 0) {
+    return [
+      `# Evaluation Runs: ${result.experimentSlug}`,
+      "",
+      "_This experiment has no runs yet._",
+      "",
+      `> Trigger a run with \`platform_run_experiment\` (slug: \`${result.experimentSlug}\`) to populate this list.`,
+    ].join("\n");
+  }
+
+  const lines: string[] = [];
+  lines.push(`# Evaluation Runs: ${result.experimentSlug}`);
+  lines.push("");
+  lines.push(
+    `Showing ${result.runs.length} of ${result.pagination.totalHits} run${result.pagination.totalHits === 1 ? "" : "s"}.`,
+  );
+  if (result.pagination.hasMore) {
+    lines.push(
+      `> More runs exist beyond this limit. Re-run with a higher \`limit\` (max ${MAX_LIMIT}) if needed.`,
+    );
+  }
+  lines.push("");
+  lines.push("| Run ID | Status | Started | Finished | Result |");
+  lines.push("|--------|--------|---------|----------|--------|");
+
+  for (const run of result.runs) {
+    lines.push(
+      `| \`${run.runId}\` | ${runStatus(run)} | ${formatEpoch(run.timestamps.createdAt)} | ${formatEpoch(run.timestamps.finishedAt ?? null)} | ${passRate(run.summary?.evaluations ?? {})} |`,
+    );
+  }
+
+  lines.push("");
+  lines.push(
+    "> Use `platform_experiment_status <runId>` for live progress, or `platform_experiment_results <runId>` for per-row results.",
+  );
+  return lines.join("\n");
+}

package/src/tools/list-experiments.ts

@@ -0,0 +1,95 @@
+import { makeRequest } from "../langwatch-api.js";
+
+interface ExperimentSummary {
+  id: string;
+  slug: string;
+  name: string | null;
+  type: string;
+  workflowId: string | null;
+  createdAt: string;
+  updatedAt: string;
+  runsCount: number;
+  lastRunAt: string | null;
+}
+
+interface ExperimentListResponse {
+  experiments: ExperimentSummary[];
+  pagination: {
+    page: number;
+    pageSize: number;
+    totalHits: number;
+    hasMore: boolean;
+  };
+}
+
+const DEFAULT_LIMIT = 25;
+const MAX_LIMIT = 100;
+
+const formatTimestamp = (iso: string | null): string => {
+  if (!iso) return "—";
+  const d = new Date(iso);
+  if (Number.isNaN(d.getTime())) return iso;
+  return d.toISOString().replace("T", " ").slice(0, 19) + " UTC";
+};
+
+const escapeMarkdownTableCell = (value: string): string =>
+  value.replace(/[\\|]/g, (c) => `\\${c}`).replace(/\n/g, " ");
+
+const formatInlineCode = (value: string): string =>
+  `\`${value.replace(/[\\`|]/g, (c) => `\\${c}`).replace(/\n/g, " ")}\``;
+
+export async function handleExperimentList(params: {
+  limit?: number;
+}): Promise<string> {
+  const requested =
+    typeof params.limit === "number" && params.limit > 0
+      ? params.limit
+      : DEFAULT_LIMIT;
+  const effectiveLimit = Math.min(requested, MAX_LIMIT);
+
+  const search = new URLSearchParams();
+  search.set("pageSize", String(effectiveLimit));
+
+  const result = (await makeRequest(
+    "GET",
+    `/api/experiments?${search.toString()}`,
+  )) as ExperimentListResponse;
+
+  if (result.experiments.length === 0) {
+    return [
+      "# Experiments",
+      "",
+      "_No experiments found in this project._",
+      "",
+      "> Create an experiment in the LangWatch dashboard, then re-run this tool to discover its slug.",
+    ].join("\n");
+  }
+
+  const lines: string[] = [];
+  lines.push("# Experiments");
+  lines.push("");
+  lines.push(
+    `Showing ${result.experiments.length} of ${result.pagination.totalHits} experiment${result.pagination.totalHits === 1 ? "" : "s"}.`,
+  );
+  if (result.pagination.hasMore) {
+    lines.push(
+      `> More experiments exist beyond this limit. Re-run with a higher \`limit\` (max ${MAX_LIMIT}) if needed.`,
+    );
+  }
+  lines.push("");
+  lines.push("| Slug | Name | Runs | Last Run |");
+  lines.push("|------|------|------|----------|");
+
+  for (const exp of result.experiments) {
+    const name = escapeMarkdownTableCell(exp.name ?? exp.slug);
+    lines.push(
+      `| ${formatInlineCode(exp.slug)} | ${name} | ${exp.runsCount} | ${formatTimestamp(exp.lastRunAt)} |`,
+    );
+  }
+
+  lines.push("");
+  lines.push(
+    "> Use `platform_experiment_list_runs` with one of these slugs to see its runs.",
+  );
+  return lines.join("\n");
+}

package/src/tools/list-projects.ts

@@ -0,0 +1,41 @@
+import { listProjects as apiListProjects } from "../langwatch-api-projects.js";
+
+export async function handleListProjects(params?: {
+  page?: number;
+  limit?: number;
+}): Promise<string> {
+  const result = await apiListProjects({
+    page: params?.page,
+    limit: params?.limit ?? 100,
+  });
+  const projects = result.data;
+
+  if (!Array.isArray(projects) || projects.length === 0) {
+    return "No projects found in this organization.\n\n> Tip: Use `platform_create_project` to create your first project.";
+  }
+
+  const lines: string[] = [];
+  lines.push(`# Projects (${result.pagination.total} total)\n`);
+
+  for (const p of projects) {
+    lines.push(`## ${p.name}`);
+    lines.push(`**ID**: ${p.id}`);
+    lines.push(`**Slug**: ${p.slug}`);
+    lines.push(`**Language**: ${p.language}`);
+    lines.push(`**Framework**: ${p.framework}`);
+    lines.push(`**Updated**: ${p.updatedAt}`);
+    lines.push("");
+  }
+
+  if (result.pagination.totalPages > 1) {
+    lines.push(
+      `> Showing page ${result.pagination.page} of ${result.pagination.totalPages}. Use \`page\` parameter to navigate.`,
+    );
+  }
+
+  lines.push(
+    "> Use `platform_get_project` with the ID to see full project details.",
+  );
+
+  return lines.join("\n");
+}

package/src/tools/revoke-api-key.ts

@@ -0,0 +1,9 @@
+import { revokeApiKey as apiRevokeApiKey } from "../langwatch-api-api-keys.js";
+
+export async function handleRevokeApiKey(params: {
+  id: string;
+}): Promise<string> {
+  await apiRevokeApiKey(params.id);
+
+  return `API key ${params.id} revoked successfully.`;
+}

package/src/tools/{run-evaluation.ts → run-experiment.ts}

@@ -23,12 +23,12 @@ interface EvaluationStatusResponse {
   };
 }
 
-export async function handleRunEvaluation(params: {
+export async function handleRunExperiment(params: {
   slug: string;
 }): Promise<string> {
   const result = (await makeRequest(
     "POST",
-    `/api/evaluations/v3/${encodeURIComponent(params.slug)}/run`,
+    `/api/experiments/${encodeURIComponent(params.slug)}/run`,
   )) as EvaluationRunResponse;
 
   const lines: string[] = [];
@@ -39,18 +39,18 @@ export async function handleRunEvaluation(params: {
   if (result.runUrl) lines.push(`**View at**: ${result.runUrl}`);
   lines.push("");
   lines.push(
-    `> Use \`platform_evaluation_status\` with run ID "${result.runId}" to check progress.`,
+    `> Use \`platform_experiment_status\` with run ID "${result.runId}" to check progress.`,
   );
 
   return lines.join("\n");
 }
 
-export async function handleEvaluationStatus(params: {
+export async function handleExperimentStatus(params: {
   runId: string;
 }): Promise<string> {
   const status = (await makeRequest(
     "GET",
-    `/api/evaluations/v3/runs/${encodeURIComponent(params.runId)}`,
+    `/api/experiments/runs/${encodeURIComponent(params.runId)}`,
   )) as EvaluationStatusResponse;
 
   const lines: string[] = [];

package/src/tools/update-project.ts

@@ -0,0 +1,22 @@
+import { updateProject as apiUpdateProject } from "../langwatch-api-projects.js";
+
+export async function handleUpdateProject(params: {
+  id: string;
+  name?: string;
+  language?: string;
+  framework?: string;
+  piiRedactionLevel?: "STRICT" | "ESSENTIAL" | "DISABLED";
+}): Promise<string> {
+  const project = await apiUpdateProject(params);
+
+  const lines: string[] = [];
+  lines.push(`Project updated successfully!\n`);
+  lines.push(`**Name**: ${project.name}`);
+  lines.push(`**ID**: ${project.id}`);
+  lines.push(`**Slug**: ${project.slug}`);
+  lines.push(`**Language**: ${project.language}`);
+  lines.push(`**Framework**: ${project.framework}`);
+  lines.push(`**PII Redaction**: ${project.piiRedactionLevel}`);
+
+  return lines.join("\n");
+}