Skip to content

mcp

ivcap mcp

Start an MCP server for accessing all tools on an IVCAP platform

Synopsis

Start an MCP (Model Context Protocol) server.

In addition to MCP Tools for calling platform services, this server also exposes the ivcap-cli release's embedded agent skills via MCP Resources and Prompts ("MCP-Provisioned Skills").

Resources (preferred if your MCP client exposes resources/read to LLMs): - skills://manifest JSON manifest of available skills - skills://catalog.json JSON catalog (metadata + hashes; no bodies) - skills://CONTEXT.md General agent best-practices for ivcap-cli - skills://SKILLS.md Top-level skills tree index - skills://{name}/SKILL.md Skill playbook body (markdown) - skills://file/{path} Any embedded markdown file (e.g. category SKILLS.md)

Genome tools: - genome_import Import an NCBI genome into IVCAP (artifact, assembly report and manifest aspects) - genome_list List the genomes imported into IVCAP

Bridge Tools (fallback for clients that don't expose resources/read): - search skills='*' lists all skill playbooks (same as skills://manifest) and skills='' finds the matching ones; tools='' searches the deployed services and the built-in tools - read_skill Read a skill playbook by name (same as skills://{name}/SKILL.md)

Prompts: - use-ivcap-best-practices Instructs an agent to load CONTEXT + relevant skills (prefers resources/read, falls back to tools if needed)

Note: Many MCP clients (including Claude Desktop) don't currently expose resources/read to LLMs for programmatic access. The bridge tools provide equivalent functionality via tools/call until client support improves.

Recommended MCP client system prompt:

Before answering any task: 1. Call resources/list on all connected MCP servers (if available) 2. Identify resources matching: SKILL.md, instructions, prompt 3. Fetch and read matching resources via resources/read (if available) 4. Otherwise use the search tool (skills='') and the read_skill tool 5. Apply those instructions when completing the user's request

ivcap mcp [flags]

Options

  -h, --help                 help for mcp
      --log-dir string       directory for MCP log files (default "/tmp")
      --port int             optional port to open for SSE connection to MCP server (default -1)
  -s, --tool-schema string   the schema URN used for describing MCP tools (default "urn:sd-core:schema.ai-tool.1")
      --with-logging         enable JSON-RPC request/response logging to file

Options inherited from parent commands

      --access-token string   Access token to use for authentication with API server [IVCAP_ACCESS_TOKEN]
      --context string        Context (deployment) to use
      --debug                 Set logging level to DEBUG
      --no-history            Do not store history
  -o, --output string         Set format for displaying output [json, yaml]
      --silent                Do not show any progress information
      --timeout int           Max. number of seconds to wait for completion (default 30)

SEE ALSO

  • ivcap - A command line tool to interact with a IVCAP deployment