By antiguru
Read Firefox-format performance profiles (samply, perf-imported) from Claude Code via an MCP server. Bundles skills for the canonical recording workflow and view-preset cookbook.
Diagnoses a broken pollard install — MCP server unreachable, missing binaries, profile load failures, unsymbolicated frames. Use when any pollard tool call fails, when load_profile errors out, or when the user reports "the profiler isn't working."
Records a performance profile with samply (or converts an existing perf.data via samply import) and loads it into pollard for analysis. Use when the user wants to profile a binary, a one-shot command, an already-running process, or analyze an existing perf.data file.
Applies canonical hide_modules / hide_frames regex sets to a loaded profile via pollard's create_view. Use when a Rust profile is dominated by framework noise — tracing-subscriber Layered walls, tokio runtime internals, or Rust stdlib glue (drop, format, panic) — and you want to focus on user code.
A Model Context Protocol (MCP) server that exposes Firefox-format performance
profiles (such as those produced by samply)
to AI coding assistants.
The name comes from pollarding — the forestry technique of pruning a tree's upper branches to encourage dense, manageable regrowth — which is exactly what this tool does to call trees so an LLM can hold them.
Published to crates.io.
pollard exposes 16 MCP tools:
load_profile, unload_profile, list_profiles,
create_view, describe_profile, summarytop_functions, top_groups, call_tree, stacks_containing,
folded_stacks, compare_profilessource_for_function, asm_for_function,
address_to_function, compare_functionstop_functions, call_tree, and compare_profiles accept an optional
event argument: omit it to aggregate the default samples track
(cycles, in samply's perf recorder) or pass a marker name like
cache-misses, branch-misses, or instructions to aggregate that
hardware counter instead.
top_groups currently aggregates samples only.
See docs/superpowers/specs/2026-04-28-pollard-design.md for full details.
See docs/superpowers/specs/2026-05-06-view-presets-cookbook.md for
copy-paste hide_modules / hide_frames regex sets covering common
Rust noise (tracing-subscriber, tokio internals, stdlib glue). The same
cookbook ships as the pollard:view-presets skill — see
.claude-plugin/ for the bundled plugin layout that registers both the
MCP server and the skills.
The pollard binary always has to be on your PATH — the plugin
bundle does not ship it. Install with cargo:
cargo install pollard
Or build the latest from this repository:
cargo install --git https://github.com/antiguru/pollard
Then pick one of two ways to register pollard with Claude Code.
Installs the MCP server and the bundled skills (profile-recording,
view-presets, pollard-doctor) in one step:
/plugin marketplace add antiguru/antiguru-tools
/plugin install pollard@antiguru-tools
The antiguru-tools marketplace also lists sibling plugins
(binutils-mcp, timely-skill, differential-skill, columnar-skill).
Skills surface as /pollard:<skill> alongside everything else the
user has installed. If a tool call fails after install, run
/pollard:pollard-doctor — the doctor walks the install in dependency
order and surfaces the exact remediation.
If you only want the tools and not the skills, register the binary directly as a user-scoped MCP server:
claude mcp add pollard pollard --scope user
Either path makes load_profile, top_functions, call_tree and
the rest of the tools listed above available in any Claude Code
session.
cargo build --release
MIT OR Apache-2.0.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
npx claudepluginhub antiguru/pollard --plugin pollardSkill for working with the columnar crate for struct-of-arrays data layout.
Harness-native ECC operator layer - 67 agents, 271 skills, 92 legacy command shims, reusable hooks, rules, selective install profiles, and production-ready workflows for Claude Code, Codex, OpenCode, Cursor, and related agent harnesses
Reliable automation, in-depth debugging, and performance analysis in Chrome using Chrome DevTools and Puppeteer
Plugin that includes the Figma MCP server and Skills for common workflows
Production-grade vibe coding suite. Structured workflows from discovery to deployment: start, build, ship, fix, refactor.
Connect to Atlassian products including Jira and Confluence. Search and create issues, access documentation, manage sprints, and integrate your development workflow with Atlassian's collaboration tools.
Memory compression system for Claude Code - persist context across sessions