{"$schema":"https://wellknown.network/schemas/agent-record-v1.json","schemaVersion":"1","id":"ag_67manhk9nznb","handle":"perfdigest-mcp","url":"https://wellknown.network/agents/perfdigest-mcp","links":{"self":"https://wellknown.network/agents/perfdigest-mcp/record.json","html":"https://wellknown.network/agents/perfdigest-mcp","markdown":"https://wellknown.network/agents/perfdigest-mcp/record.md","api":"https://wellknown.network/api/v1/agents/perfdigest-mcp","status":"https://wellknown.network/api/v1/agents/perfdigest-mcp/status","claim":"https://wellknown.network/agents/perfdigest-mcp/claim","claimApi":"https://wellknown.network/api/v1/claims","claimDescriptor":"https://wellknown.network/agents/perfdigest-mcp/claim.json","badge":"https://wellknown.network/agents/perfdigest-mcp/badge.svg","openapi":"https://wellknown.network/openapi.json","history":"https://wellknown.network/api/v1/agents/perfdigest-mcp/history","tools":"https://wellknown.network/api/v1/agents/perfdigest-mcp/tools"},"ard":{"identifier":"urn:air::server:perfdigest-mcp","type":"application/mcp-server-card+json"},"kind":"mcp_server","declared":{"name":"perfdigest-mcp","summary":"Local MCP server that digests development-loop reports — profilers (NVIDIA ncu, AMD rocprof, Linux perf, Apple Metal, chrome-trace), build tools (ptxas, clang -ftime-trace, cmake, ninja, cargo), benchmarks (criterion), CI run logs, and git change state — into token-efficient JSON for LLM coding age…","description":"# perfdigest\n\nLocal **MCP server** that makes development-loop reports **token-efficient** for\nLLM coding agents — profiler output first, and since v1.2 the rest of the loop's\nfeedback artifacts too (build traces, build diagnostics, benchmark results, CI\nrun logs, repo change state). It sits between the tool and the agent: reads the\nreport from disk and returns a small, structured, numeric digest the agent can\nact on, while keeping a pointer back to the raw report for lazy expansion.\n\nIt is a **translator/router, not a judge** — interpretation (\"is this kernel\nmemory-bound?\") is the model's job. perfdigest only provides efficient,\ndeterministic access to clean numeric metrics across vendors and languages.\n\n## Two operations — keep them separate\n\nperfdigest deliberately splits what an agent does with a profiler into two tiers:\n\n1. **Digest (read a report) — universal, available on every machine.** A report's\n   origin is irrelevant to digesting it. An NVIDIA `.ncu-rep` (or its CSV export)\n   captured on a CI/remote GPU runner can be pulled to a **Mac** and digested\n   there — that is how you do CUDA work on a Mac via CI. Tier-1 is never gated by\n   local hardware, only by whether a reader's dependency imports.\n2. **Capture (produce a report) — platform-verified.** You cannot run `ncu` on a\n   Mac, Metal on Linux, or hardware PMU counters under WSL2. `platform_capabilities`\n   and `suggest_profile_command` gate this so the agent never spends context on a\n   capture that can't run here, and redirect it to \"capture elsewhere, digest here.\"\n\n## The four feedback channels (v1.2 — the Development Observatory)\n\nEvery change a session makes propagates through the same four stages, each with\nits own verbose artifact: *what changed* (RepoState) → *does it build, at what\ncost* (BuildDigest) → *does it pass* (CIDigest) → *how fast is it* (PerfDigest).\nperfdigest digests all four through the **same seven tools** — a backend is a\nregistry row, not a new API surface. One principle…","publisher":{"name":"Hamza Usta","url":null},"homepage":null,"repository":"https://github.com/onlyxItachi/PerfDigest-MCP/issues","version":"1.3.0","license":null,"protocols":["mcp"],"tags":["claude","codex","cuda","llm","mcp","metal","nsight","perf","profiler","rocm"],"pricing":null,"endpoints":[{"url":"pypi:perfdigest-mcp","type":"package_pypi","auth":null,"probeable":false}],"skills":null,"tools":null,"extra":null,"attribution":{"kind":"pypi","name":"pypi","repoUrl":"pypi","summary":"pypi","version":"pypi","description":"pypi","publisherName":"pypi"}},"derived":{"capabilities":[{"slug":"ai.evaluation","name":"Evaluation & Benchmarks","confidence":0.802,"provenance":"derived"}],"categories":["ai"],"language":"en"},"observed":{"status":"unknown","statusReason":"Distributed as a package to run locally; no network endpoint to check.","lastOkAt":null,"lastProbedAt":null,"statusComputedAt":null,"reliability30d":null,"latestObservations":[],"tools":null,"package":{"name":"perfdigest-mcp","registry":"pypi","observedAt":"2026-10-28T19:22:50.667Z","publishedAt":"2026-10-12T11:18:47.630172Z","latestVersion":"1.3.0"},"toolSurface":null,"endpointFacts":[]},"verification":{"claimed":false,"claimedAt":null,"proofs":[]},"provenance":{"sources":[{"source":"pypi","key":"perfdigest-mcp","url":"https://pypi.org/project/perfdigest-mcp/","firstSeenAt":"2026-09-10T09:26:01.165Z","fetchedAt":"2026-10-28T19:21:50.283Z","normalizedAt":"2026-10-28T19:21:50.283Z"}]},"firstSeenAt":"2026-09-10T09:26:01.165Z","updatedAt":"2026-10-28T19:22:50.667Z"}