---

name: agentic-and-mcp description: "OmicVerse CLI, MCP server, registry manifest, JARVIS/gateway, smart agent, agent backend, skill seeker, and AI-assisted session workflows." disable-model-invocation: true metadata: disco-role: operating license: GPL 3.0

Agentic and MCP

Use this sub-skill for OmicVerse command-line entry points, the MCP server, registry/manifest inspection, JARVIS and gateway workflows, smart-agent configuration, provider backends, ov-skill-seeker, and AI-assisted notebook/session execution.

For package-wide routing and installation strategy, return to ../../SKILL.md.

When to Use

• Inspect command-line routing for omicverse, omicverse claw, omicverse jarvis, omicverse gateway, omicverse web, omicverse skill-seeker, omicverse-mcp, or ov-skill-seeker.
• Configure an MCP-compatible client to discover OmicVerse registry tools through stdio or local streamable-http transport.
• Inspect registry manifests with omicverse.mcp.build_default_manifest(...), omicverse.mcp.get_manifest(...), or bundled scripts/inspect_registry.py.
• Use or debug JARVIS/gateway, one-shot claw -q, channel credentials, OpenAI/Gemini/Ollama/provider configuration, Codex/Gemini CLI OAuth, or web-launcher behavior.
• Use omicverse.utils.smart_agent.Agent(...), OmicVerseAgent, AgentConfig, streaming events, approval policy, notebook execution, filesystem context, or harness traces.
• List, validate, package, or create bundled agent skills with ov-skill-seeker.

Route Elsewhere

• Core AnnData IO, QC, preprocessing, plotting, report/provenance, and registry discovery from Python: ../core-analysis/SKILL.md.
• Single-cell annotation, integration, marker, trajectory, fate, communication, or perturbation analysis: ../single-cell-workflows/SKILL.md.
• Bulk, enrichment, metabolomics, proteomics, microbiome, and table statistics: ../multiomics-statistics/SKILL.md.
• Spatial, histology, deconvolution, mapping, and spatial IO workflows: ../spatial-integration/SKILL.md.
• Genetics, AIRR, molecular, raw alignment, external binaries, and downloads: ../specialist-domains/SKILL.md.

Safe First Checks

Run help and read-only introspection before starting services or contacting providers:

python sub-skills/agentic-and-mcp/scripts/check_mcp_runtime.py --profile fast-mock
python sub-skills/agentic-and-mcp/scripts/inspect_registry.py --phase P0+P0.5 --limit 20
python sub-skills/agentic-and-mcp/scripts/inspect_registry.py --search pca --phase P0+P0.5 --show-schema
omicverse --help
omicverse-mcp --help
ov-skill-seeker --help

Expected signals:

• check_mcp_runtime.py prints JSON with package versions, executable availability, MCP public API flags, and CLI help flags.
• inspect_registry.py prints JSON with tool_count, selected tools, execution-class/category counts, and no network listener.
• omicverse-mcp --help lists --phase, --transport, --session-id, --persist-dir, --max-adata, --max-artifacts, --host, --port, and --http-path.

Service-Gated Commands

Do not start long-running services as a default validation step. Start these only when the user explicitly asks for a service or MCP client integration:

omicverse-mcp --phase P0+P0.5 --transport stdio
python -m omicverse.mcp --phase P0 --transport stdio
python -m omicverse.mcp --phase P0+P0.5 --transport streamable-http --host 127.0.0.1 --port 8765 --http-path /mcp
omicverse claw -q "show code to QC this AnnData object"
omicverse gateway --web-host 127.0.0.1 --web-port 5050 --no-browser

Use stdio MCP only with clients that keep stdout reserved for JSON-RPC. Use local streamable-http only when a network listener is acceptable and the user has approved host/port binding.

References

• references/mcp-and-cli.md: console scripts, MCP phases/transports, manifests, meta tools, session/persistence limits, and skill seeker commands.
• references/agent-runtime.md: JARVIS/gateway, smart agent factory, AgentConfig, provider/auth modes, streaming, notebook/session behavior, and permission policy.
• references/troubleshooting.md: missing extras, stdio pollution, startup timeouts, registry hydration, credentials, persistence/quota errors, gateway/web imports, and sandbox policy.