[aterga]

Summary

Follow-up to #26. Adds a real Internet Identity ↔ MCP connect-handshake test: a genuine Internet Identity canister (the release-2026-07-17 wasm) runs in PocketIC and issues a real canister-signed registration delegation, which the real McpServer redeems through its own ic-agent. Nothing about II is mocked — every canister interaction and every signature is genuine. The test plays only the browser/frontend role (the human's clicks).

It also lands the supporting production change that makes this possible: verify the registration chain against the injected agent's root key rather than a hardcoded mainnet key, so a PocketIC-hosted II's chain verifies (and, in production, the agent's key is mainnet — behavior-preserving).

Related issues

Follow-up to #26.

Changes

• src/e2e_handshake.rs (new) — in-crate test, #[cfg(all(test, feature = "e2e"))]. Starts PocketIC, installs the real II wasm with the captcha disabled (init arg), then:
  • plays the frontend: register an anchor → mcp_set_config (trust the server) → prepare_mcp_registration_delegation → get_mcp_registration_delegation (the real canister-signed P_reg → X hop);
  • drives the real server end to end: /oauth/register → /oauth/authorize (mints X) → /oauth/connect/redeem (the server signs mcp_register_v2 on the live II, verifying the canister signature against PocketIC's fetched root key) → /oauth/token → asserts the minted bearer token authenticates (session_for_token) and its TTL tracks the II grant.
  • II candid types are hand-rolled from the release-tag .did (no git dependency).
• src/identities.rs — registration_identity now uses DelegatedIdentity::new_with_root_key(&self.agent.read_root_key()) instead of DelegatedIdentity::new (hardcoded mainnet). Ties chain-trust to the same agent that makes the mcp_register_v2 call; the reply is already certificate-verified against that key, so this is consistent, not a weakening.
• Cargo.toml — pocket-ic as an optional dependency behind a new e2e feature; the default build compiles and downloads neither.
• Removed the interim cfg(test) registration seam and its mocked connect→token test — redeem_registration_delegation now has a single real path.

Testing

• cargo build --locked --all-targets — default build stays lean (no pocket-ic).
• cargo test --locked --all-targets — 136 lib + 7 router tests pass.
• cargo test --features e2e --no-run — the PocketIC test compiles.
• Executing the e2e test needs two artifacts cargo does not fetch — II_WASM (the release internet_identity_backend.wasm.gz) and POCKET_IC_BIN (the pocket-ic v15 server binary) — and could not be run in this environment (release-asset downloads are blocked here). Run it where those are provisioned: II_WASM=… POCKET_IC_BIN=… cargo test --features e2e. The test skips cleanly when they're unset, so default CI stays green; a dedicated CI job that downloads them (and later bumps the wasm) is the follow-up.
• cargo fmt/clippy: the repo is not default-rustfmt-clean by design and CI gates on build+test only; the new code matches the surrounding style and adds no clippy warnings.

Checklist

• I have read the Contributing guidelines.
• Docs/comments updated — the new module documents the flow and exactly how to run it.
• No secrets, credentials, or internal-only information are included.

🤖 Generated with Claude Code

https://claude.ai/code/session_017h4prMHkLo6UnwnuoPgmUj

[Copilot]

Pull request overview

Adds a test-only seam to stub the non-hermetic Internet Identity registration redemption step, enabling a new fully hermetic end-to-end unit test that exercises the MCP↔II connect handshake success path from connect/redeem through OAuth token exchange and token-authenticated session resolution.

Changes:

• Introduces a #[cfg(test)] Identities::with_register_override hook to override redeem_registration_delegation’s live/crypto-dependent registration step while preserving real session bookkeeping (grant expiry + permissions).
• Adds test_store_registering plus a new test that drives connect_redeem → token → session_for_token, including PKCE S256 enforcement, TTL bounded by grant expiry, and single-use authorization codes.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/identities.rs	Adds a cfg(test)-only registration override seam for hermetic tests, without affecting non-test builds.
src/auth.rs	Adds a hermetic success-path test (redeem → token → session auth) and a helper store wired to the new seam.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

Comments suppressed due to low confidence (2)

src/auth.rs:2825

• The new hermetic success-path test seeds AuthzPending directly but never mints the connect’s registration key (X). In production, redemption requires that X was minted for the connect, and the delegation chain’s final hop must target it. Consider minting X via store.identities.registration_pubkey_b64(state).await and passing its DER bytes into well_formed_reg_chain_json(...) to keep the test exercising those invariants.

        // A PKCE pair, seeded as an in-flight connect whose initiator cookie the
        // redeem must present (bypassing the browser redirect that would set it).
        let verifier = "connect-redeem-verifier-0123456789";
        let challenge = super::pkce_s256(verifier);
        let cookie = "bind-connect";
        let state = "sess-connect";
        store.authz.write().await.insert(

src/auth.rs:2853

• After minting x_der, pass it into well_formed_reg_chain_json(&x_der) so the redeem request’s chain targets the connect’s registration key (X).

            headers,
            axum::Json(super::RedeemBody {
                state: state.into(),
                delegation: well_formed_reg_chain_json(),
            }),

[Copilot]

Pull request overview

Copilot reviewed 5 out of 6 changed files in this pull request and generated 1 comment.

Comments suppressed due to low confidence (1)

Cargo.toml:61

• This feature comment also references tests/e2e_pocket_ic.rs, but the gated PocketIC test is src/e2e_handshake.rs. Updating the path here will keep the run instructions discoverable and accurate.

# Compiles + enables the real II<>MCP handshake test against a live Internet
# Identity canister in PocketIC (needs II_WASM + POCKET_IC_BIN at runtime; see
# tests/e2e_pocket_ic.rs). Off by default: the standard `cargo test` neither
# builds `pocket-ic` nor touches the network.

[Copilot]

Pull request overview

Copilot reviewed 5 out of 6 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 5 out of 6 changed files in this pull request and generated 1 comment.

[Copilot]

Pull request overview

Copilot reviewed 5 out of 6 changed files in this pull request and generated no new comments.

Comments suppressed due to low confidence (2)

src/e2e_handshake.rs:277

• These doc comments claim the OAUTH_CLIENTS_FILE override “can't leak into other test threads running in parallel”, but the env var is process-global and will be visible to any concurrently running test while this one is executing (it’s only restored on drop). This can make --features e2e test runs flaky when other tests also call SharedClients::load() (e.g. tests/routers.rs).

/// Restores (or clears) the process-global `OAUTH_CLIENTS_FILE` on drop — even on
/// a panic-unwind — so this test's override can't leak into other test threads
/// running in parallel, and removes the temp file it pointed at.

src/e2e_handshake.rs:310

• Same issue as above: this comment says the env-var override “can't leak into other test threads”, but it can during the test run. Consider updating the wording and/or ensuring the recommended invocation runs this test in isolation (e.g. filter by test name and/or use --test-threads=1).

    // Isolate the OAuth client store: a UNIQUE temp file (so concurrent runs
    // never collide on a fixed name) plus a guard that restores/clears the
    // process-global `OAUTH_CLIENTS_FILE` on drop — even on a panic — so the
    // override can't leak into other test threads.