Skip to main content
Glama

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault
HW_NATIVE_SYS_ROOTNoPath to the hw-native-sys workspace root. Optional; if not set, the server uses config/repos.json's 'workspace_root' field (default '..') or falls back to project_root().parents[1]...

Instructions

Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.

This server publishes no instructions, or was last inspected before Glama recorded them.

Capabilities

Features and capabilities supported by this server

Protocol revision2025-11-25

CapabilityDetails
tools
{
  "listChanged": false
}
prompts
{
  "listChanged": false
}
resources
{
  "subscribe": false,
  "listChanged": false
}
experimental
{}

Tools

Functions exposed to the LLM to take actions

NameDescription
list_repositoriesA

List configured repositories, their paths, and architecture metadata.

repository_healthA

Get branch, dirty state, and ahead/behind for every configured repo.

search_codeB

Search code across repos. mode=locations (default) omits line text; mode=context includes context.

list_tasksA

List available named tasks for a repository, including metadata and warnings.

run_taskA

Run a configured task in a repository.

Build/test tasks are routed by NPU availability: with NPUs reachable they run on the host (allowed by policy); without NPUs they are re-routed into the sim Docker image for the repo (see _SIM_RUN), or refused with guidance when the repo has no sim image (e.g. PTOAS).

run_commandA

Run an ad-hoc shell command in a repository. Prefer run_task for known tasks or git_log/git_diff/read_file for read-only operations.

When no NPU is reachable, build/test commands (cmake, make, ninja, pip install, pytest, …) are refused — those must run inside the sim Docker images, never on the local repo. Read-only commands are unaffected.

explain_taskB

Show the exact command and metadata configured for a named task.

git_logA

Return the last N commits for a repository as structured data (sha, author, date, message).

git_diffA

Return the git diff for a repository (stat + optional patch). Prefer stat_only=True for orientation.

read_fileA

Read an arbitrary source file from a repository without shelling out.

list_task_typesA

Return all valid task_type values with one-line descriptions for use with bootstrap_session and route_task.

route_taskB

Return read-first docs, rules, entrypoints, and verify tasks for a compiler workflow.

list_knowledge_topicsA

List available task routes, MCP resources, notes topics, and bootstrap prompts.

read_docA

Read a workspace document with tier labeling. Optional section extracts a markdown heading.

explain_abstractionA

Explain a stack abstraction: IR/passes/codegen/ISA/runtime or Ascend hardware (AIC, HCCL, etc.).

search_abstractionsA

Search the abstraction index by keyword. Results are ranked by relevance (exact name > name-contains > tag > layer/kind).

explain_passA

Explain a pass in the Default pipeline: order, phase, neighbors, verify tasks.

program_statusA

Structured PR/plan status from status_prs.md (open PRs, blockers, plan cross-index).

collective_statusA

Collective-comm feature parity status (merged/planned/gap) from the parity matrix in pypto-3.0-notes/distributed/current_status.md. Read-only -- never writes to the source doc.

verify_ladderA

Suggest minimal verify tasks for a set of changed file paths.

Returns static_checks (e.g. ["clang-tidy"]) when a changed path is a C/C++ file in a C++ repo — run those before the pytest tasks (see tools/clang_tidy_workflow).

summarize_profileA

Summarize a pypto-profiling campaign directory (results.json, anomalies).

trace_contractA

Full cross-layer trace: stack location + contract triangle + cross-layer verify tasks + active-PR links.

This is the enriched trace. Use trace_in_stack for a lightweight stack-location-only lookup.

find_entrypointsB

Find code entrypoints for a repo and optional area (e.g. pypto, codegen_orch).

trace_in_stackA

Lightweight stack location: matched abstraction card or path-prefix pipeline stage only.

No contract artifacts or PR links. Use trace_contract for the enriched cross-layer trace.

knowledge_healthA

Check knowledge config health: missing paths, stale enriched docs, Ascend corpus, index build time.

ascend_env_checkA

Read-only Ascend/CANN environment check: devices, HCCL preload, Docker hints.

generate_verify_handoffA

Generate markdown handoff for developer NPU verification in a container.

bootstrap_sessionB

Single-call session bootstrap: route metadata, read_plan, health, program hints.

gate_pr_scriptB

Generate a bash script for the safe rebase->pre-commit->sim-Docker-tests->squash->push gate workflow.

Prompts

Interactive templates invoked by user choice

NameDescription
start_compiler_work
start_distributed_work
start_ascend_work
start_npu_verify
finish_work

Resources

Contextual data attached and managed by the client

NameDescription
resource_pytorch-hccl-tests_overview
resource_pytorch-hccl-tests_bandwidth-runbook
resource_overview_ecosystem
resource_overview_pipeline
resource_pypto_ir
resource_pypto_passes
resource_pypto_codegen
resource_pypto_distributed
resource_ptoas_overview
resource_pto-isa_overview
resource_simpler_overview
resource_simpler_l3_distributed_collectives
resource_pypto-lib_overview
resource_pypto-lib_status
resource_pypto-lib_building_blocks
resource_pypto-lib_attention
resource_pypto-lib_models
resource_pypto-lib_distributed_support
resource_pypto-lib_moe
resource_agent_invariants
resource_agent_distributed_work_policy
resource_flows_compile_to_device
resource_flows_matmul_demo
resource_flows_distributed_allreduce
resource_flows_dependency_triangle
resource_flows_performance
resource_ascend_hardware
resource_ascend_arch_families
resource_ascend_memory_hierarchy
resource_ascend_cann_mapping
resource_ascend_hccl_runtime
resource_ascend_platform_decisions
resource_ascend_alignment_rules
resource_ascend_hccl_container_checklist
resource_tools_sim_docker_workflow
resource_tools_clang_tidy_workflow
resource_tools_gate_pr_workflow
resource_notes_freshness
resource_notes_stack_availability
resource_notes_abstractions_master
resource_notes_pypto_abstractions
resource_notes_simpler_abstractions
resource_notes_ptoisa_abstractions
resource_notes_ptoas_abstractions
agent_routing_resource
notes_pypto_lib_status
notes_pypto_lib_building_blocks
notes_pypto_lib_attention
notes_pypto_lib_models
notes_pypto_lib_distributed_support
notes_moe
notes_pass_infrastructure
notes_codegen_infrastructure
notes_dependency_triangle
notes_notes_simpler
notes_kernel_orchestration
notes_host_collectives
notes_distributed_work_policy
notes_runtime_arch_index
notes_runtime_design
notes_runtime_async
notes_multi_level_runtime_ring
notes_machine_hierarchy
notes_simpler_distributed_runtime_design
notes_serving_implementation_plan
notes_sharded_tensor
notes_tensor_layout
notes_tensor_valid_shape
notes_tpush_tpop_isa_design

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/georgebisbas/mcp-hw-native-sys'

If you have feedback or need assistance with the MCP directory API, please join our Discord server