mcp-agent
Build, compose, serve, operate, and troubleshoot mcp-agent applications, workflows, MCP servers, Temporal workers, CLI/cloud deployments, and observability integrations.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- mcp-agent
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Build, compose, serve, operate, and troubleshoot mcp-agent applications, workflows, MCP servers, Temporal workers, CLI/cloud deployments, and observability integrations.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/repo-provenance.md2.6 KB
- 参考文档references/repo-routing-metadata.json496 B
- 参考文档references/troubleshooting.md4.4 KB
- 脚本scripts/check_mcp_agent_environment.py5.0 KB
03 SKILL.md 正文
mcp-agent
Use this repo skill when a task involves the mcp-agent Python package or its CLIs: building MCP-enabled agents, composing effective workflow patterns, exposing applications as MCP servers, operating MCP Agent Cloud, using Temporal durability, or configuring logging/tracing/integrations.
First Checks
- Confirm the package is available with
python -c "import mcp_agent; print(mcp_agent.__name__)"or install it withpip install "mcp-agent"/uv add "mcp-agent". - Add optional extras only for the selected workflow:
mcp-agent[openai],[anthropic],[google],[azure],[bedrock],[cohere],[temporal],[langchain],[crewai], or[redis]. - Keep secrets in
mcp_agent.secrets.yaml, environment variables, or deployment secret handles; keep reusable defaults inmcp_agent.config.yamlor aSettingsobject. - Read
references/repo-provenance.mdbefore deciding whether this skill matches a checkout or whetherrefresh-repo-skillis needed. - Read
references/troubleshooting.mdfor cross-cutting install, import, config, optional dependency, CLI, and service failures.
Route by Task
- Core SDK and app code: use
sub-skills/core-sdk/SKILL.mdforMCPApp,Agent,AgentSpec,AugmentedLLM,RequestParams, settings, secrets, local function tools, decorators, and factory basics. - Workflow composition: use
sub-skills/workflow-patterns/SKILL.mdfor routers, intent classifiers, parallel fan-out/fan-in, orchestrators, deep orchestrators, evaluator-optimizer loops, swarms, and structured outputs. - MCP server integration: use
sub-skills/mcp-server-integration/SKILL.mdfor upstream MCP server config, transports, roots, sampling, elicitation, OAuth/auth,create_mcp_server_for_app, and app-as-server validation. - CLI and Cloud operations: use
sub-skills/cli-cloud-operations/SKILL.mdformcp-agent,mcp-cloud,mcpc, project scaffolding, diagnostics, local dev commands, deploy, logs, env/secrets, workflows, and client install. - Durable execution: use
sub-skills/durable-execution/SKILL.mdwhen workflows need Temporal workers, long-running tools, pause/resume, signals, retry policies, activity registration, or service-backed durability. - Observability and integrations: use
sub-skills/observability-integrations/SKILL.mdfor logger transports, OpenTelemetry, token accounting, progress/streaming, provider extras, LM Studio/local endpoints, LangChain, and CrewAI adapters.
Common Decision Points
- Start with
core-sdkwhen the user asks to create or debug Python application code; route toworkflow-patternsonly after the app/agent basics are clear. - Route to
mcp-server-integrationwhen the words are about MCP transports, server config, app-as-server behavior, OAuth, resources, prompts, sampling, roots, elicitation, or client/server tool surfaces. - Route to
cli-cloud-operationswhen the task is command syntax, scaffolding, config checks, deployments, logs, hosted app state, environment secrets, or installing a server into Claude Code/Cursor/VS Code/ChatGPT clients. - Route to
durable-executionwhen the user already has or wants a workflow that must survive restarts, pause for human input, receive signals, or run in a Temporal worker. - Route to
observability-integrationswhen failures involve provider wrapper imports, OpenTelemetry exporters, token counters, JSONL event logs, streaming events, local model endpoints, or LangChain/CrewAI adapters.
Safe Validation Helpers
sub-skills/core-sdk/scripts/check_core_sdk.pychecks imports, signatures, local tools, decorators,AgentSpec, andRequestParamswithout credentials or network.sub-skills/workflow-patterns/scripts/check_workflow_imports.pyverifies canonical workflow imports and class names such asLLMRouter.sub-skills/mcp-server-integration/scripts/validate_server_config.pychecks MCP server config shape and optional executable availability.sub-skills/cli-cloud-operations/scripts/collect_cli_help.pycollects safe--helpsnapshots without deploy/login/install side effects.sub-skills/durable-execution/scripts/check_temporal_config.pyvalidates Temporal config statically without starting a worker.sub-skills/observability-integrations/scripts/check_observability_config.pyandsub-skills/observability-integrations/scripts/summarize_event_log.pyinspect logger/OTEL config and JSONL event logs safely.
Safety Notes
- Do not run Cloud deployment, client install writes, env secret mutation, workflow resume/cancel, or public server transports unless the user explicitly intends mutation.
- Do not assume optional provider SDKs, Temporal, Redis, LangChain, CrewAI, or cloud credentials are present in a base install; install and validate the smallest extra needed.
- Do not paste API keys or secret handles into generated code, logs, support transcripts, or client snippets.
- Prefer read-only
--help, config validation, bundled checker scripts, import checks, and dry-run/static modes before starting services or calling external providers.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).