Transports

urirun keeps the URI contract separate from the transport. The same URI can be called locally, through a service endpoint, or by a flow orchestrator.

Local and shell

• local-function calls an in-process function registered by code.
• argv-template renders an argv list and executes it without a shell.
• shell-template renders a shell string and requires explicit policy approval.

Docker

Docker examples use URI targets as service names:

python://python-worker/text/normalize
node://node-worker/text/slugify
shell://shell-worker/report/write

See v2/examples/docker_uri_flow for a Compose flow where services publish bindings and an orchestrator runs a multi-step URI flow.

HTTP and browser

The HTML example in v2/examples/html_uri_app loads a binding document, renders URI forms, and calls a Python backend through POST /api/run.

The backend can expose logs, recent calls, MCP tools, and A2A cards from the same registry, so frontend actions use the same URI names as backend actions.

The connector hub also has HTTP-oriented packages. The verified HTTP Check connector installs from connect.ifuri.com/connectors/http-check and exposes:

httpcheck://host/http/query/status

That route can be called from shell, an ifuri host process, an MCP client or a flow orchestrator because the transport-specific implementation remains behind the registry adapter.

gRPC

urirun.v2_grpc provides a small RPC surface for route listing, unary calls, and stream-style calls. Install the optional dependency set when using it:

pip install "urirun[grpc] @ git+https://github.com/tellmesh/urirun.git@v0.3.12#subdirectory=adapters/python"

MCP and A2A

Because v2 bindings include JSON Schema, the registry can be projected into:

• MCP tools/list
• MCP tools/call
• A2A agent card skills

Execution still goes through the same urirun policy gate.

See MCP and A2A for the LLM-facing projection.