refactor: simplify use_sentry tool by removing constraint parameters (#599)

Remove organizationSlug, projectSlug, and regionUrl parameters from
use_sentry tool. The tool now uses context constraints directly without
merging user-provided overrides.

Also add benchmark-agent.sh script for comparing direct vs agent mode
performance, and remove UserPromptSubmit hook from Claude settings.

---------

Co-authored-by: Claude <noreply@anthropic.com>

Author: dcramer

.claude/settings.json

@@ -21,19 +21,6 @@
     ],
     "deny": []
   },
-  "hooks": {
-    "UserPromptSubmit": [
-      {
-        "matcher": "*",
-        "hooks": [
-          {
-            "type": "command",
-            "command": "echo 'MANDATORY: If something is unclear, you MUST ask me. ALWAYS reference our docs when they are available for a task, and make a note when they arent.'"
-          }
-        ]
-      }
-    ]
-  },
   "enableAllProjectMcpServers": true,
   "includeCoAuthoredBy": true,
   "enabledMcpjsonServers": ["sentry"]

AGENTS.md

@@ -67,7 +67,7 @@ pnpm run generate-otel-namespaces  # Update OpenTelemetry docs
 
 # Manual Testing (preferred for testing MCP changes)
 pnpm -w run cli "who am I?"                    # Test with local dev server (default)
-pnpm -w run cli --agent "who am I?"            # Test agent mode (use_sentry tool)
+pnpm -w run cli --agent "who am I?"            # Test agent mode (use_sentry tool) - approximately 2x slower
 pnpm -w run cli --mcp-host=https://mcp.sentry.dev "query"  # Test against production
 pnpm -w run cli --access-token=TOKEN "query"   # Test with local stdio mode
 

benchmark-agent.sh

@@ -0,0 +1,110 @@
+#!/bin/bash
+
+# Benchmark script for comparing direct vs agent mode performance
+# Usage: ./benchmark-agent.sh [iterations]
+
+ITERATIONS=${1:-10}
+QUERY="what organizations do I have access to?"
+
+echo "=========================================="
+echo "MCP Agent Performance Benchmark"
+echo "=========================================="
+echo "Query: $QUERY"
+echo "Iterations: $ITERATIONS"
+echo ""
+
+# Arrays to store timings
+declare -a direct_times
+declare -a agent_times
+
+echo "Running direct mode tests..."
+for i in $(seq 1 $ITERATIONS); do
+    echo -n "  Run $i/$ITERATIONS... "
+
+    # Run and capture timing (real time in seconds)
+    START=$(date +%s.%N)
+    pnpm -w run cli "$QUERY" > /dev/null 2>&1
+    END=$(date +%s.%N)
+
+    # Calculate duration
+    DURATION=$(echo "$END - $START" | bc)
+    direct_times+=($DURATION)
+
+    echo "${DURATION}s"
+done
+
+echo ""
+echo "Running agent mode tests..."
+for i in $(seq 1 $ITERATIONS); do
+    echo -n "  Run $i/$ITERATIONS... "
+
+    # Run and capture timing
+    START=$(date +%s.%N)
+    pnpm -w run cli --agent "$QUERY" > /dev/null 2>&1
+    END=$(date +%s.%N)
+
+    # Calculate duration
+    DURATION=$(echo "$END - $START" | bc)
+    agent_times+=($DURATION)
+
+    echo "${DURATION}s"
+done
+
+echo ""
+echo "=========================================="
+echo "Results"
+echo "=========================================="
+
+# Calculate statistics for direct mode
+direct_sum=0
+direct_min=${direct_times[0]}
+direct_max=${direct_times[0]}
+for time in "${direct_times[@]}"; do
+    direct_sum=$(echo "$direct_sum + $time" | bc)
+    if (( $(echo "$time < $direct_min" | bc -l) )); then
+        direct_min=$time
+    fi
+    if (( $(echo "$time > $direct_max" | bc -l) )); then
+        direct_max=$time
+    fi
+done
+direct_avg=$(echo "scale=2; $direct_sum / $ITERATIONS" | bc)
+
+# Calculate statistics for agent mode
+agent_sum=0
+agent_min=${agent_times[0]}
+agent_max=${agent_times[0]}
+for time in "${agent_times[@]}"; do
+    agent_sum=$(echo "$agent_sum + $time" | bc)
+    if (( $(echo "$time < $agent_min" | bc -l) )); then
+        agent_min=$time
+    fi
+    if (( $(echo "$time > $agent_max" | bc -l) )); then
+        agent_max=$time
+    fi
+done
+agent_avg=$(echo "scale=2; $agent_sum / $ITERATIONS" | bc)
+
+# Calculate difference
+diff=$(echo "scale=2; $agent_avg - $direct_avg" | bc)
+percent=$(echo "scale=1; ($agent_avg - $direct_avg) / $direct_avg * 100" | bc)
+
+echo ""
+echo "Direct Mode:"
+echo "  Min:     ${direct_min}s"
+echo "  Max:     ${direct_max}s"
+echo "  Average: ${direct_avg}s"
+echo ""
+echo "Agent Mode:"
+echo "  Min:     ${agent_min}s"
+echo "  Max:     ${agent_max}s"
+echo "  Average: ${agent_avg}s"
+echo ""
+echo "Difference:"
+echo "  +${diff}s (${percent}% slower)"
+echo ""
+
+# Show all individual results
+echo "All timings:"
+echo "  Direct: ${direct_times[*]}"
+echo "  Agent:  ${agent_times[*]}"

docs/testing.mdc

@@ -88,7 +88,7 @@ Interactive testing with the MCP test client (preferred for testing MCP changes)
 # Test with local dev server (default: http://localhost:5173)
 pnpm -w run cli "who am I?"
 
-# Test agent mode (use_sentry tool only)
+# Test agent mode (use_sentry tool only) - approximately 2x slower
 pnpm -w run cli --agent "who am I?"
 
 # Test against production

packages/mcp-cloudflare/src/client/components/fragments/remote-setup.tsx

@@ -68,7 +68,8 @@ export default function RemoteSetup() {
           <strong>Agent Mode:</strong> Reduce context by exposing a single{" "}
           <code>use_sentry</code> tool instead of individual tools. The embedded
           AI agent handles natural language requests and automatically chains
-          tool calls as needed.
+          tool calls as needed. Note: Agent mode approximately doubles response
+          time due to the embedded AI layer.
         </p>
         <ul>
           <li>

packages/mcp-server/src/tools/use-sentry/handler.test.ts

@@ -113,47 +113,6 @@ describe("use_sentry handler", () => {
     expect(toolNames).toHaveLength(19);
   });
 
-  it("handles requests with organization slug parameter", async () => {
-    mockUseSentryAgent.mockResolvedValue({
-      result: {
-        result: "Found errors in my-org",
-      },
-      toolCalls: [],
-    });
-
-    const result = await useSentry.handler(
-      {
-        request: "Find errors in my-org",
-        organizationSlug: "my-org",
-      },
-      mockContext,
-    );
-
-    expect(mockUseSentryAgent).toHaveBeenCalled();
-    expect(result).toBe("Found errors in my-org");
-  });
-
-  it("handles requests with project slug parameter", async () => {
-    mockUseSentryAgent.mockResolvedValue({
-      result: {
-        result: "Issues in frontend project",
-      },
-      toolCalls: [],
-    });
-
-    const result = await useSentry.handler(
-      {
-        request: "Show issues in frontend project",
-        organizationSlug: "my-org",
-        projectSlug: "frontend",
-      },
-      mockContext,
-    );
-
-    expect(mockUseSentryAgent).toHaveBeenCalled();
-    expect(result).toBe("Issues in frontend project");
-  });
-
   it("wraps tools with session constraints", async () => {
     const constrainedContext: ServerContext = {
       ...mockContext,

packages/mcp-server/src/tools/use-sentry/handler.ts

@@ -1,14 +1,8 @@
 import { z } from "zod";
-import { setTag } from "@sentry/core";
 import { InMemoryTransport } from "@modelcontextprotocol/sdk/inMemory.js";
 import { experimental_createMCPClient } from "ai";
 import { defineTool } from "../../internal/tool-helpers/define";
 import type { ServerContext } from "../../types";
-import {
-  ParamOrganizationSlug,
-  ParamRegionUrl,
-  ParamProjectSlug,
-} from "../../schema";
 import { useSentryAgent } from "./agent";
 import { buildServer } from "../../server";
 import { serverContextStorage } from "../../internal/context-storage";
@@ -58,9 +52,6 @@ export default defineTool({
       .describe(
         "The user's raw input. Do not interpret the prompt in any way. Do not add any additional information to the prompt.",
       ),
-    organizationSlug: ParamOrganizationSlug.optional(),
-    projectSlug: ParamProjectSlug.optional(),
-    regionUrl: ParamRegionUrl.optional(),
     trace: z
       .boolean()
       .optional()
@@ -73,26 +64,6 @@ export default defineTool({
     openWorldHint: true,
   },
   async handler(params, context: ServerContext) {
-    // Set tags for monitoring
-    if (params.organizationSlug) {
-      setTag("organization.slug", params.organizationSlug);
-    }
-    if (params.projectSlug) {
-      setTag("project.slug", params.projectSlug);
-    }
-
-    // Create context with updated constraints from user parameters
-    // This ensures the embedded agent respects org/project constraints
-    const contextWithConstraints: ServerContext = {
-      ...context,
-      constraints: {
-        organizationSlug:
-          params.organizationSlug || context.constraints.organizationSlug,
-        projectSlug: params.projectSlug || context.constraints.projectSlug,
-        regionUrl: params.regionUrl || context.constraints.regionUrl,
-      },
-    };
-
     // Create linked pair of in-memory transports for client-server communication
     const [clientTransport, serverTransport] =
       InMemoryTransport.createLinkedPair();
@@ -101,15 +72,15 @@ export default defineTool({
     // eslint-disable-next-line @typescript-eslint/no-unused-vars
     const { use_sentry, ...toolsForAgent } = tools;
 
-    // Build internal MCP server with constrained context
+    // Build internal MCP server with the provided context
     const server = buildServer({
-      context: contextWithConstraints,
+      context,
       tools: toolsForAgent,
     });
 
     // Run all MCP operations within the ServerContext
     // This ensures tools invoked through the MCP protocol have access to the context
-    return await serverContextStorage.run(contextWithConstraints, async () => {
+    return await serverContextStorage.run(context, async () => {
       // Connect server to its transport
       await server.server.connect(serverTransport);