Prax & the interoperability frontier

Where Prax sits in the emerging agent-interoperability ecosystem — what it already does, and the frontier it can move onto. (Goal: "Prax becomes part of the interoperability frontier.")

Synced from Prax at f62d7985 View source ↗

← Architecture

Where Prax sits in the emerging agent-interoperability ecosystem — what it already does, and the frontier it can move onto. (Goal: “Prax becomes part of the interoperability frontier.”)

Note on scope. This was prompted by Pangram Space, which turned out to be an interpretability project (visualizing an AI-text- detector’s activations), not an interoperability standard — there’s nothing there to join. It belongs (if anywhere) to the LLM-behavior/interpretability lane, not here. But the goal it surfaced is real, so this doc maps Prax against it.

The five axes of agent interoperability

  1. Expose — let other agents/products use your capabilities.
  2. Consume — use other agents’/services’ capabilities.
  3. Discover — find capabilities across orgs/the web, and verify them.
  4. Interchange — move knowledge/artifacts between systems in a portable format.
  5. Identity & provenance — authenticate who/what produced a request or a piece of content (credentials, signing, content authentication).

Where Prax is today [verified]

Axis Prax today
Expose ✅ Strong. The MCP server (prax/mcp/, infrastructure/mcp-server.md) exposes a curated, bearer-gated, per-caller-identity subset of Prax tools to other agents over JSON-RPC MCP. Governance stays in front; write tools grantable per-caller, HIGH never. This is the “make Prax usable by other agents” surface.
Consume ⚠️ Gap. Prax is an MCP server, not yet an MCP client — it can’t yet treat another agent’s MCP server (or an A2A peer) as a tool source. prax/mcp/clients.py is the caller registry, not an outbound client.
Discover ◻️ Assessed, deferred. Google’s ARD (research/agentic-resource-discovery.md) — publish/discover/verify capabilities via ai-catalog.json — verdict was document-don’t-adopt-now (cross-org web discovery isn’t a single-user harness’s problem; its trust half is redundant with Prax’s imported→HIGH + capability gateway).
Interchange ✅ Shipped. The OKF bridge (knowledge_export_okf/import, research/open-knowledge-format.md) makes the concept graph round-trippable to portable markdown bundles.
Identity & provenance ⚠️ Partial. Per-caller MCP identity + ephemeral capability TTLs exist (inbound). No outbound content provenance — Prax doesn’t sign/attest the content it generates across channels.

The frontier (ranked by fit)

  1. MCP-as-client / A2A (the “Consume” axis) — highest-value, most natural. Prax is hub-and-spoke; an outbound MCP/A2A client makes any external agent or MCP server a spoke — Prax consumes others’ capabilities the same way it exposes its own. This is the cleanest “join the frontier” move and reuses the delegation pattern. (Tool-overload still applies — gate behind the on-demand tool-search / hub-and-spoke discipline.) Tracked: ../IDEAS_BACKLOG.md #27.
  2. Content provenance / credentials (the “Identity & provenance” axis). Prax produces content across Discord/SMS/TeamWork/notes. Participating in content- authentication standards (C2PA Content Credentials, signed attestations of “produced by Prax/agent X”) is a real interop frontier — and the honest flip side of the AI-detection world Pangram lives in (interpretability shows AI text is reliably detectable → aim to be verifiably attributable, not undetectable; ../research/pangram-detector-interpretability.md). Lets downstream systems verify provenance instead of guessing. Tracked: ../IDEAS_BACKLOG.md #28.
  3. Capability discovery (the “Discover” axis) — when cross-org matters. ARD (or a successor) becomes relevant if Prax ever needs to discover third-party capabilities at web scale rather than being handed endpoints. Deferred until there’s a multi-org use case; the assessment is already on file.

Verdict

Prax is already a first-class interop provider (MCP server + OKF interchange) — that half of the frontier is shipped. The gap is the consumer half: an outbound MCP/A2A client is the single highest-leverage step to “part of the interoperability frontier,” because it closes the loop (expose and consume) using machinery Prax already has (hub-and-spoke, governed tools, per-caller identity). Content provenance is the natural second frontier. Capability discovery (ARD) stays deferred until a cross-org need is concrete.

See also: infrastructure/mcp-server.md, research/open-knowledge-format.md, research/agentic-resource-discovery.md.