natprof-mcp
macOS native profiling and debugging MCP server that wraps sample, atos, vmmap, leaks, and lldb to enable an agent to profile, symbolicate, and debug running or freshly launched processes without manual path or port handling.
Warning. Serious findings were identified. Review the full report before connecting. Grades reflect the full trust pyramid: code, verification depth, and reputation. New projects cap at C until adoption is earned.
How we verified
Verified is a snapshot. Live keeps it current, and builds your track record.
⚡ Connect GitHub → continuous verification on every pushwhy connect →Who stands behind it
Ronin15
Source: Glama
Claim it to get a verified publisher badge, a free copy of our full audit findings, and direct contact for any high-priority issues we find. Or connect your repo for our deepest verification, Live Monitored: read-only, revoke anytime. What we access →
Install from
The grade above is for the source repository. Registries can serve a different version, so we mark the ones we were not able to read.
These names and descriptions are the publisher's own, read from the source code. We print them as written. Our assessment is the findings above, not this list.
open_sessionOpen a session on a running process. Captures ASLR slides and LC_UUIDs, and snapshots every non-system image. After this call you may rebuild freely — symbolication uses the frozen copies.
launch_sessionLaunch a fresh binary under LLDB, instead of attaching to a pid that's already running. This is the only way to set environment variables before the target's first instruction executes — open_session() only ever attaches to something already past that point, so MallocStackLogging can't be turned on …
debugRun any LLDB command: `bt`, `thread list`, `frame variable`, `expr`, `breakpoint set -n foo`, `watchpoint`, `memory read`.
backtrace_allInterrupt, dump every thread's backtrace, resume.
light_backtraceLow-impact "what's every thread doing" snapshot via `sample`, instead of an LLDB `process interrupt`. The process is never fully halted — `sample` suspends one thread at a time just long enough to read its stack — so this is safe to run against latency-sensitive processes (audio, rendering, networki…
recordStart a stack-sampling profile via `sample`, in the background — this returns immediately rather than blocking the caller for `seconds`. MCP tool calls are otherwise synchronous: a client with no way to background a call would be stuck fully blocked for the whole recording, with no visibility into p…
record_statusCheck on a record() that was started in the background. Returns {status: "running"} while `sample` is still collecting, or {status: "done", report} once the trace is ready for hotspots() — report is the path to the raw `sample` output, not a parsed result; hotspots() is what actually reads it. Raise…
hotspotsSymbolicated self-time hotspots from the last completed record(). System frames excluded by default — they dominate leaf counts and say nothing. Symbols are UUID-verified or the call fails; there is no stale-symbol path.
set_baselineSnapshot the top 100 current hotspots as the comparison baseline for a later compare() call. Internally calls hotspots(top_n=100), so it needs a completed record() first — same "call record_status() until done" requirement, and it raises the same errors hotspots() would if no trace exists yet.
compareDiff the baseline set by set_baseline() against a fresh hotspots() reading taken right now — so a new record()/record_status() cycle must complete after set_baseline() and before calling this, or you're just comparing the baseline against itself. Keyed on image!symbol, since two dylibs can both expo…
leaksLeak check. LeakSanitizer's at-exit detection does not work on Darwin arm64, so `leaks` is the route on Apple Silicon.
verifyRe-check every snapshotted image against its recorded LC_UUID, and every dSYM against its image. Run this if you suspect the symbols.
close_sessionDetach LLDB, delete snapshots and trace, drop the session.
Tool annotations
No tools have read-only/destructive annotations
Add readOnlyHint or destructiveHint annotations to every tool so hosts can warn users before invoking.
All four hints declared on every tool
13/13 tools missing one or more hints — open_session (missing: readOnlyHint, destructiveHint, idempotentHint, openWorldHint); launch_session (missing: readOnlyHint, destructiveHint, idempotentHint, openWorldHint); debug (missing: readOnlyHint, destructiveHint, idempotentHint, openWorldHint), +10 more. OpenAI's directory rejects tools where any of the four hints are missing or non-boolean.
For every tool, set all four hints (readOnlyHint, destructiveHint, idempotentHint, openWorldHint) to explicit true/false values that match the handler’s actual behaviour.
Destructive tools are labelled
1 tool perform destructive updates without destructiveHint — close_session deletes at line 978 (shutil.rmtree(s.workdir(), ignore_errors=True))
Add destructiveHint:true to any tool whose handler calls .delete(), .upsert(), .update(), unlink, rm, DELETE, DROP, REPLACE INTO, or any operation that overwrites existing data.
Descriptions match behaviour
1 tool describes read intent but its handler mutates — leaks (line 915: out = subprocess.run(["leaks", str(s.pid), "--list"],)
Rename the tool, rewrite the description, or move the side-effect into a separate clearly-named tool.
Tool handlers catch errors
Only 1/13 tool handlers wrap calls in try/catch (8%)
Wrap each tool handler body in try/catch and return a structured error response.
License file
No license file
Add a LICENSE file (MIT, Apache-2.0, etc.).
Tests exist
No test files found
Add tests that exercise each declared tool.
Tool description accuracy
leaks: description implies read-only but handler writes/deletes/executes
Update tool descriptions to accurately reflect all capabilities — especially write, delete, or execute operations.
Claim the listing to review these findings one by one and send us a correction where you disagree, straight to the team. Claiming also means we tell you when the grade moves, and reach you first if we find anything urgent.
[](https://m8ven.ai/mcp/ronin15/natprof-mcp)?variant=verified from the URL.Vetting this one by hand? Tool Check is an MCP that scores other MCPs. Add it once and ask Claude, ChatGPT, or any MCP client to grade a server, surface CVEs, check the publisher, and suggest safer alternatives — before you install.
https://m8ven.ai/api/mcp/tool-check