Perfetto MCP Server

Community

Turn natural language into powerful Perfetto trace analysis

How to install ↓ Compare GitHub source repository ↗

Category: Dev Tools #74 of 438
Stars: ★ 221 Popular
Transport: stdio (local process)
Runtime: Python 3.13+
Credentials: No credential needed
License: Apache-2.0
Last commit: 6 months ago ⚠
Tools: 12

63FMRS · C

A comprehensive Android performance-analysis MCP server that turns natural-language prompts into 12 specialized Perfetto tools covering ANR, CPU, frame performance, contention, Binder IPC, and memory analysis. Stdio transport only; requires Python 3.13+ and uv; community-maintained under Apache-2.0, not official.

Strongest · Documentation 15/20 Weakest · Reliability 8/20

Reliability

8/20

Security and permissions

14/20

Maintenance

12/20

Documentation

15/20

Setup experience

14/20

Why each score

Reliability 8/20: The README lists 12 tools (find_slices, execute_sql_query, detect_anrs, etc.) and describes an output format (Summary/Details/Metadata), but the repository provides no verifiable CI workflow or committed tests demonstrating the MCP init + tool-list handshake, error handling, or that declared tools match real behavior. With no source code, tests, or CI evidence supplied, judgment rests solely on self-reported README claims, and per the static calibration rule this caps the score low. The happy path is plausible on paper but unproven, and edge cases (trace_processor download failures, malformed SQL, missing trace files) are undocumented.
Security and permissions 14/20: The server performs local, read-only analysis of user-supplied Perfetto trace files via SQL queries; there are no visible irreversible destructive defaults (delete/pay/deploy) and no hardcoded credentials or overt exfiltration in the README. One external network behavior exists: by default the underlying Perfetto Python library may download a trace_processor_shell binary, and the project mitigates this by offering PERFETTO_MCP_TRACE_PROCESSOR_BIN_PATH to point to a local binary in restricted networks — a reasonable, documented mitigation. However, the README does not clearly disclose whether any trace data (which may contain sensitive app internals) ever leaves the local machine, nor does it define explicit permission/data-flow boundaries, so it falls short of full marks.
Maintenance 12/20: The project carries an Apache-2.0 license, is not archived, and shows 214 stars with only 5 open issues, suggesting some baseline activity. However, the provided material contains no evidence of commit cadence, release/version history, a changelog, dependency-update practices, or a security-response channel, and no CI badge or release notes are visible. This limits the score to a middling value reflecting apparent-but-unverified activity rather than clearly governed, sustained maintenance.
Documentation 15/20: The README is well organized: prerequisites (Python 3.13+, uv), detailed setup snippets for four mainstream clients (Cursor, Claude Code, VS Code, Codex), fallback install paths (pip, local dev via uv), a categorized tool table with purpose and example prompts, an optional env-var override, and links to external Perfetto docs — solid layered documentation. It falls short of full marks because detailed parameter schemas are missing (only two required params are documented; optional filters are described loosely), and there is no troubleshooting section, no explicit limits, and no cost disclosure.
Setup experience 14/20: The README provides concrete one-click install links and JSON config examples for major clients (Cursor, Claude Code, VS Code, Codex) plus pip and local-dev (uv sync + uv run) fallbacks, keeping the process low-friction. However, since this is a purely static review with no CI workflow or committed tests to confirm these install paths actually work end-to-end, the calibration rule caps setup at 15; the score reflects a clear, well-documented process short of verified execution proof.

Static review · not runListed 2026-09-07

Read the FMRS scoring method →

Fit and risk

What it can accessReads local filesRuns commands or code

Best for

Android developers analyzing Perfetto traces
Engineers who want jank, ANR, contention, and memory analysis without writing SQL
Clients supporting stdio MCP such as Cursor, Claude Code, VS Code, and Codex

Not for

Users who do not analyze Perfetto traces
Scenarios where an absolute trace_path and process_name cannot be provided
Environments that cannot run Python MCP servers via uvx or pip

Required permissions

Needs read access to local Perfetto trace files (absolute path passed as trace_path)
May download trace_processor_shell on first run (or use a local binary via PERFETTO_MCP_TRACE_PROCESSOR_BIN_PATH)

Risks and side effects

Traces contain application and system runtime details; sharing them may expose sensitive information
Default behavior may download an external binary; restricted networks should configure a local path
Queries are generated by an LLM and results should be verified

Setup

Before you start

Runtime：Python 3.13+

Other optional settings (1)

PERFETTO_MCP_TRACE_PROCESSOR_BIN_PATH optional Optional absolute path to a local trace_processor_shell binary; set only if your network blocks downloads, otherwise leave unset.

Requires Python 3.13+ and uv (on macOS: brew install [email protected], brew install uv). Add the MCP server to Cursor, Claude Code, VS Code, or Codex with command 'uvx perfetto-mcp'. Alternatively pip3 install perfetto-mcp and run python3 -m perfetto_mcp. If your network blocks downloads, set PERFETTO_MCP_TRACE_PROCESSOR_BIN_PATH to the absolute path of a local trace_processor_shell binary. For development, run uv sync and uv run mcp dev src/perfetto_mcp/dev.py inside the perfetto-mcp-server directory.

.cursor/mcp.json

{"mcpServers":{"perfetto-mcp":{"command":"uvx","args":["perfetto-mcp"]}}}

Shown for Cursor. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json

{
  "servers": {
    "perfetto-mcp": {
      "command": "uvx",
      "args": [
        "perfetto-mcp"
      ]
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal

claude mcp add perfetto-mcp -- uvx perfetto-mcp

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

After install, confirm tools like find_slices and detect_anrs appear in your client's tool list; then ask "Use perfetto trace /path/to/trace.perfetto-trace for process com.example.app to detect ANRs" and a structured JSON answer proves the connection works.

Troubleshooting

Verify trace_path is an absolute path to a .perfetto-trace file
Confirm process_name matches the process in the trace (e.g. com.example.app)
If trace_processor_shell download fails, set PERFETTO_MCP_TRACE_PROCESSOR_BIN_PATH to a local binary
Ensure Python 3.13+ and uv are installed
Be explicit in prompts, e.g. 'Use perfetto trace /absolute/path for process com.example.app'

Things to try

Once connected, you can ask your AI assistant things like:

Use perfetto trace /absolute/path/to/trace.perfetto-trace for process com.example.app and detect ANRs with severity summary
List main-thread hotspot slices >50 ms during 10s–25s
Find janky frames above 16.67 ms and list the worst 20
Analyze heap dominator classes and list top memory offenders

Tools 12

find_slices read-only

Survey slice names and locate hot paths

execute_sql_query read-only

Run custom PerfettoSQL for advanced analysis

detect_anrs read-only

Find ANR events with severity classification

anr_root_cause_analyzer read-only

Deep-dive ANR causes with ranked likelihood

cpu_utilization_profiler read-only

Profile thread-level CPU usage and scheduling

main_thread_hotspot_slices read-only

Find longest-running main thread operations

detect_jank_frames read-only

Identify frames missing deadlines

frame_performance_summary read-only

Summarize overall frame health metrics

Show 4 more tools

thread_contention_analyzer read-only

Find synchronization bottlenecks

binder_transaction_profiler read-only

Analyze Binder IPC performance

memory_leak_detector read-only

Find sustained memory growth patterns

heap_dominator_tree_analyzer read-only

Identify memory-hogging classes

Use cases

Explain why FragmentManager takes 438ms to execute

Detect and analyze ANR root causes

Find CPU hot threads and main-thread hotspot slices

Detect janky frames above 16.67ms and summarize frame performance

Locate lock contention and slow Binder transactions

Detect memory-leak signals and analyze heap dominator classes

Supported clients

Cursor

Claude Code

VS Code

Codex

Listed from the project's documentation, not tested by this site.

Overview

Perfetto MCP Server is a Model Context Protocol (MCP) server that transforms natural-language prompts into focused Perfetto analyses. It explains jank, diagnoses ANRs, spots CPU hot threads, uncovers lock contention, and finds memory leaks without writing SQL. Every tool requires trace_path (absolute path to a Perfetto trace) and process_name (target process/app name). All tools return structured JSON with summary, details, and metadata.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 63.3k · Tools 2 Compare with this →

NPM Sentinel MCP Server 79 · B

AI-powered NPM package analysis MCP server

★ 18 · Tools 19 Compare with this →

MartinLoop MCP Server 79 · B

An independent control layer that gives AI coding agents budgets, brakes, and receipts

★ 413 Compare with this →

GitHub MCP Server 76 · B

GitHub's official MCP server for managing repos, issues, PRs, and workflows via natural language

★ 33.7k · Tools 15 Compare with this →

Compare with other MCP servers

Source revision 174d14570b62 Data synced 2026-10-19 Read the FMRS scoring method