feat(apr-code): thread .mcp.json env field to subprocess spawn (PMAT-CODE-MCP-ENV-001)#1574
Merged
Conversation
…CODE-MCP-ENV-001)
Closes PMAT-CODE-MCP-ENV-001 — the P2 follow-up from
PMAT-CODE-MCP-JSON-LOADER-001. Previously the `env` field on
.mcp.json's `mcpServers` entries was parsed (the JSON shape was
deny-unknown-fields safe) but DISCARDED at the
McpServerEntry → McpServerConfig conversion step.
Three-layer plumbing:
1. **manifest.rs** — `McpServerConfig` gains `env: BTreeMap<String, String>`
(with `#[serde(default)]` so existing TOML manifests stay valid). Both
the struct and `McpTransport` derive `Default` so test literals can
use `..Default::default()` if they prefer.
2. **mcp_json.rs** — `entry_to_config` now copies `entry.env.clone()`
into the produced `McpServerConfig.env` (single-line fix at the
step that was previously dropping the field).
3. **tool/mcp_client.rs** — `StdioMcpTransport` gains an `env` field +
a new `new_with_env(server, command, env)` constructor.
`send_jsonrpc` (the spawn site shared by `call_tool` and
`discover_tools`) now applies env via `tokio::process::Command::env`
for each (k, v) pair. The existing `new(...)` constructor is
preserved for back-compat — it just yields an empty env map.
`register_mcp_client_tools` updated to call `new_with_env(...)` so the
manifest's env reaches the spawned subprocess in the live happy path.
## Tests
* 4 new mcp_json tests:
- `parse_entry_env_is_collected` — env shape parses unchanged
- `entry_to_config_threads_env` — env flows through conversion
- `entry_to_config_preserves_empty_env` — no-env yields empty map
- `merge_threads_env_to_manifest_servers` — end-to-end JSON →
AgentManifest.mcp_servers[].env
* 3 new mcp_client_tests:
- `test_stdio_transport_new_starts_with_empty_env` — back-compat
constructor
- `test_stdio_transport_new_with_env_carries_pairs` — new constructor
stores env
- `test_stdio_transport_env_flows_to_subprocess` — LIVE smoke
spawning `sh -c 'printf ... "$MCP_ENV_TEST_KEY"'` and verifying
the env var reached the spawned process. CI-skipped (subprocess
timing unreliable) but passes locally.
* All existing mcp_json (26 total now) + mcp_client (29 total now)
tests still pass — the McpServerConfig literal-construction sites
in `runtime_tests_guards.rs` got `env: Default::default()` added.
## Doc updates
`McpServerEntry.env` doc comment updated to reflect that it's no
longer "parsed but not applied" — fixed pointer to make the next
operator who reads it see the correct status.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Closes PMAT-CODE-MCP-ENV-001 — the P2 follow-up from PMAT-CODE-MCP-JSON-LOADER-001 (which noted
envwas parsed but not applied). This PR plumbs it end-to-end.Three-layer plumbing
manifest.rsMcpServerConfiggainsenv: BTreeMap<String,String>(serde-default; back-compat)mcp_json.rsentry_to_configcopiesentry.env.clone()into the config (was previously dropping it)tool/mcp_client.rsStdioMcpTransport::new_with_env;send_jsonrpcapplies viaCommand::env(k,v); oldnew(...)preservedregister_mcp_client_toolsupdated to callnew_with_env. BothMcpServerConfigandMcpTransportnow deriveDefaultso test literals can use..Default::default().Tests
sh -c 'printf ... "$MCP_ENV_TEST_KEY"'and verifying the env var arrivedMcpServerConfigliteral sites updated withenv: Default::default()Doc updates
McpServerEntry.envdoc comment fixed: was "parsed but not applied — tracked in PMAT-CODE-MCP-ENV-001 (P2)"; now "Threaded through to subprocess spawn ... PMAT-CODE-MCP-ENV-001 (CLOSED 2026-05-07)".Test plan
cargo test -p aprender-orchestrate --lib agent::mcp_json --features agents-mcp— 26/26 passcargo test -p aprender-orchestrate --lib agent::tool::mcp_client --features agents-mcp— 29/29 pass (incl. LIVE subprocess smoke)cargo build -p aprender-orchestrate --features agents-mcp— cleancargo fmt -p aprender-orchestrate -- --checkclean🤖 Generated with Claude Code