## Why
MCP inventory can be cached or collected separately from a thread's live
connections, so tool availability alone does not describe the current runtime
state.
## What changed
- Add a nullable `runtimeStatus` to `mcpServerStatus/list` for thread-scoped
requests, covering not-started, starting, connected, authentication-required,
failed, cancelled, and disabled connections.
- Observe published connection state without starting or reconnecting servers,
and return an unknown status when no thread is supplied or the active
configuration no longer matches the published registration.
- Show connection state and tool counts in the compact `/mcp` view while
retaining the detailed inventory in `/mcp verbose` and compatibility with
servers that omit `runtimeStatus`.
## Testing
- Cover runtime status transitions, deferred and disabled servers, closed
transports, configuration changes, protocol compatibility, and TUI rendering.
GitOrigin-RevId: e3bb6efe652f0fa8b3c97d5c53e4729b3a87cd91
## What changed
- Preserve valid image content from completed `node_repl` Code Mode responses alongside text evidence for Guardian reviews.
- Enable multimodal evidence for models that require automatic `node_repl` review, or when both `guardian_enhanced_node_repl_transcripts` and `guardian_node_repl_transcript_images` are enabled.
- Bound retained and rendered evidence, deduplicate images already present in review history, and fall back to text-only evidence when the reviewer lacks image support or prompt capacity.
- Keep review evidence out of the parent turn input.
## Testing
- Cover feature gating, mixed text and image ordering, truncation and eviction, unsupported reviewers, prompt pressure, image detail normalization, and reused review sessions.
GitOrigin-RevId: a3e482402f7b3484401b231dadc662e6e3504fdf
## What changed
- Capture accepted, successful `node_repl` results from Code Mode and include them as bounded, untrusted evidence in Guardian review prompts.
- Enable the enhanced transcript through `guardian_enhanced_node_repl_transcripts` or when the selected model requires automatic Node REPL review.
- Keep this evidence out of the parent model history, exclude encrypted and failed results, and avoid resending admitted evidence when a Guardian session is reused.
- Clear retained evidence and invalidate review sessions when a thread is rolled back.
## Testing
- Add unit coverage for evidence ordering, escaping, truncation, and empty responses.
- Add an integration test covering feature-disabled, feature-enabled, and model-required behavior, including filtering and review-session reuse.
GitOrigin-RevId: edc3dca5d5b88d472a492f35531ec46889a89d72
## Why
A local stdio MCP server can exit while its descendants remain alive. Cleaning up
only the server process can therefore leak processes after client shutdown.
## What changed
- Contain locally launched MCP servers in non-breakaway Windows job objects and
terminate the job during shutdown so descendants exit with the server.
- Preserve compatibility when job creation or assignment is unavailable by
retrying without containment and terminating an owned process handle instead.
- Use the process-group member fallback when terminating piped processes on
macOS.
## Testing
- Cover descendant cleanup and breakaway prevention for both local MCP protocol
modes on Windows.
- Cover the Windows nested-job assignment fallback and verify the suspended
process resumes before handle-based termination.
GitOrigin-RevId: 33d3cc122de8f7ce0ae69a649b12ea0ff64b7c6f
## Why
MCP catalog discovery must not allow a server to keep pagination running or
growing without bounds.
## What changed
- Apply the shared pagination collector to tool, resource, and resource-template
discovery.
- Limit each catalog to 100 pages and 1,024 items, and reject pagination cursors
larger than 64 KiB or any repeated cursor.
- Bound the entire pagination operation by the configured tool timeout, falling
back to 30 seconds when no timeout is configured.
## Testing
Add unit coverage for every limit and an end-to-end MCP test that preserves
valid multi-page tools while excluding a server that returns an oversized
cursor.
GitOrigin-RevId: 961a1bfa5fe8a265daf56e342a919c980cd720fa
## What changed
- Update the Rust MCP SDK from `3.0.0-beta.3` to `3.0.0` and adapt to its renamed metadata and server discovery types.
- Accept discovery responses without server identity metadata, using the configured server name as a fallback, and rely on the SDK's native support for namespaced server identity metadata.
- Preserve typed OAuth HTTP errors so transport failures, cross-origin redirects, and transient HTTP responses are reported instead of being treated as anonymous access.
## Testing
- Cover modern discovery with namespaced or missing server identity over HTTP, SSE, and stdio.
- Cover OAuth discovery error propagation for transport failures, redirects, and transient status codes.
GitOrigin-RevId: 12c1e45136cca89ce4fb15986c2b5df14608682a
## What changed
- Limit model-facing MCP namespace descriptions to 1,000 bytes without splitting UTF-8 characters or mutating the stored server metadata.
- Cap the aggregate source-description list in `tool_search` at 4 KiB while retaining source names and discovery instructions.
- Cover direct and deferred MCP tool exposure, including multibyte descriptions and tool visibility.
GitOrigin-RevId: 92736ec39475a1d6b3b03f5e3063e28225d8a5f6
## What changed
- Add an opt-in `mcp_2026_07_28` protocol mode while preserving the legacy
lifecycle by default.
- Negotiate the new protocol over streamable HTTP with `server/discover`,
including bounded responses, redirect protection, and fallback only when a
response establishes that the endpoint is legacy-only.
- Require stdio servers to opt in with
`CODEX_MCP_PROTOCOL_VERSION=2026-07-28`, and add a bounded local stdio
transport for the modern lifecycle.
- Consume paginated tool, resource, and resource-template catalogs in modern
mode, reject repeated cursors, and retain discovered server identity.
- Reconnect reusable MCP clients when their selected protocol mode changes.
## Testing
- Cover HTTP JSON and SSE discovery, legacy fallback and rejection cases,
redirects, retries, response limits, and pagination.
- Cover local and executor stdio discovery, protocol markers, message limits,
and legacy compatibility.
GitOrigin-RevId: f6a78816e127d2a482292d63b91c8384f1595903
## What changed
- Update the `rmcp` model and transport integrations for `3.0.0-beta.3`.
- Preserve legacy `elicitation/create` form requests, schema defaults, metadata, and wire-format compatibility while adopting the new elicitation types.
- Harden OAuth discovery by using GET-first discovery without starting an MCP session, preventing cross-origin header redirects, validating authorization-server and callback issuers, and retaining compatibility with metadata that omits an issuer.
- Keep stored OAuth credentials readable and avoid requiring reauthorization after transient refresh failures.
## Testing
- Add coverage for legacy elicitation round trips, OAuth discovery variants, callback issuer validation, credential migration, and refresh failures.
GitOrigin-RevId: 756197d26f3fd347c28c400228f6b3d06ed493b7
## What changed
- Convert MCP text content marked with `_meta["codex/encryptedContent"] = true`
into `encrypted_content` function call output items.
- Prefer the content-item payload whenever encrypted content is present so that
`structured_content` does not replace the encrypted output.
## Testing
- Add a stdio MCP integration test that round-trips mixed plaintext and
encrypted tool output.
GitOrigin-RevId: c3997a16b06ad38ff22b21016ede35b97974d9ec
## Why
Starting a new session should not have to wait for an unchanged stdio MCP
server to initialize before its tools can be presented to the model.
## What changed
- Cache recent tool catalogs for matching stdio MCP server configurations and
use them while a new live connection starts in the background.
- Bound the process-wide cache to 32 entries with a 30-minute TTL, and key
entries by the server configuration, environment, working directory, and
elicitation capabilities that affect the catalog.
- Keep connection-specific instructions and tool annotations out of cached
catalogs. Resolve tool calls against the live connection so tools that are
unavailable or hidden in the new session cannot be invoked from stale data.
- Publish only the newest completed catalog fetch and do not cache HTTP servers
or stdio configurations that depend on remotely sourced environment values.
## Testing
- Add coverage for cache expiration, environment changes, catalog sanitization,
and out-of-order refreshes.
- Add an end-to-end test that starts inference from cached definitions while a
replacement MCP process initializes, then verifies calls and visibility
against the live process.
GitOrigin-RevId: 2ed0f29ba53ace12532730f951b175c588aa3ae7
## Why
Prototype whether the harness can invoke the `mcp_history` MCP while
constructing full initial context and expose its thread hint to the
model without requiring a model-issued tool call.
The prototype builds on the context-window lineage added by #29256 and
is now based directly on `main`.
## What changed
- Call `mcp_history/thread_hint` with no arguments while building the
full `<token_budget>` context.
- Pass the current `threadId` through MCP request metadata, matching the
normal MCP tool-call path.
- Serialize only the unstructured `content` result and append it inside
`<token_budget>` when the call succeeds.
- Omit the additional context when the MCP call or content serialization
fails.
## Prototype limitations
- The direct call bypasses the normal model-initiated MCP approval,
lifecycle-event, telemetry, and result-sanitization path.
- The call has no prototype-specific timeout, result-size cap, or
per-window cache.
- MCP latency is added to full-context construction, including
applicable compaction paths.
## Validation
- `just test -p codex-core token_budget`
WIll make it easier to uprev when the new draft spec is supported.
Also updates reqwest where needed for compatibility but doesn't update
it everywhere since this is already a large diff.
The new version of rmcp handles certain kinds of authentication failures
differently, this patch includes support for identifying the failing scope
in a WWW-Authenticate header.
## Summary
- Treat MCP tools with `readOnlyHint: true` as parallel-safe even when
`supports_parallel_tool_calls` is unset or `false`.
- Keep server-level `supports_parallel_tool_calls` as an additive
override for non-read-only tools.
- Add focused unit coverage for the MCP handler eligibility decision.
- Update RMCP integration coverage to keep the serial baseline on a
mutable tool, verify read-only concurrency without server opt-in, and
preserve the server opt-in concurrency path separately.
## Testing
- `just fmt`
- `cargo test -p codex-core --lib tools::handlers::mcp::tests::`
- `cargo test -p codex-core --test all
stdio_mcp_read_only_tool_calls_run_concurrently_without_server_opt_in`
- `cargo test -p codex-core --test all
stdio_mcp_parallel_tool_calls_opt_in_runs_concurrently`
- `cargo test -p codex-rmcp-client`
## Why
MCP servers can provide `instructions` that explain what their tools are
for. Directly exposed MCP namespaces already use those instructions when
a connector description is not available, but deferred `tool_search`
results did not preserve that fallback. The direct path falls back from
connector metadata to server instructions, while the deferred path only
carried `connector_description` and otherwise fell back to generic
namespace text.
That meant a plain MCP server could provide useful model-facing guidance
and still appear as `Tools in the X namespace.` whenever it was
discovered lazily through `tool_search`.
## What changed
- Store one model-facing `namespace_description` on `ToolInfo`, using
connector descriptions for connector-backed tools and server
instructions for plain MCP servers.
- Thread that namespace description through the `tool_search` source
list, search indexing, and returned namespace metadata.
- Add an end-to-end regression test for deferred non-app MCP search
results exposing server instructions as the namespace description.
## Verification
- `cargo test -p codex-tools
search_tool_description_lists_each_mcp_source_once --lib`
- `cargo test -p codex-core --test all
tool_search_uses_non_app_mcp_server_instructions_as_namespace_description`
## Why
Several bug reports describe thread shutdown (including subagent
threads) leaving stdio MCP server processes behind. These reports all
point at the same lifecycle gap: Codex launches stdio MCP servers, but
the session-level shutdown path does not explicitly close MCP clients or
terminate the server process tree.
Fixes#12491Fixes#12976Fixes#18881Fixes#19469
## History
This is best understood as a regression/coverage gap in MCP session
lifecycle management, not as stdio MCP cleanup being absent all along.
#10710 added process-group cleanup for stdio MCP servers, but that
cleanup only runs when the `RmcpClient`/transport is dropped. The older
reports (#12491 and #12976) came after that cleanup existed, which
suggests the remaining problem was that some higher-level shutdown paths
kept the MCP manager alive or replaced it without explicitly draining
clients. The newer reports (#18881 and #19469) exposed the same family
around manager replacement and shutdown.
## What changed
- Added an explicit stdio MCP process handle in `codex-rmcp-client` so
local MCP servers terminate their process group and executor-backed MCP
servers call the executor process terminator.
- Added `RmcpClient::shutdown()` and manager-level MCP shutdown draining
so session shutdown, channel-close fallback, MCP refresh, and connector
probing stop owned MCP clients.
- Added regression coverage that starts a stdio MCP server, begins an
in-flight blocking tool call, shuts down the client, and asserts the
server process exits.
## Verification
- `cargo test -p codex-rmcp-client`
- `cargo test -p codex-mcp`
- `just fix -p codex-rmcp-client`
- `just fix -p codex-mcp`
- `just fix -p codex-core`
- Manual before/after validation with a temporary repro script:
- Pre-fix binary from `HEAD^` (`fed0a8f4fa`): reproduced the leak with
surviving MCP server and child PIDs, `survivors=[77583, 77592]`,
`leaked=true`.
- Post-fix binary from this branch (`67e318148b`): verified both MCP
processes were gone after interrupting `codex exec`, `survivors=[]`,
`leaked=false`.
## Summary
- Add the executor-backed RMCP stdio transport.
- Wire MCP stdio placement through the executor environment config.
- Cover local and executor-backed stdio paths with the existing MCP test
helpers.
## Stack
```text
o #18027 [6/6] Fail exec client operations after disconnect
│
@ #18212 [5/6] Wire executor-backed MCP stdio
│
o #18087 [4/6] Abstract MCP stdio server launching
│
o #18020 [3/6] Add pushed exec process events
│
o #18086 [2/6] Support piped stdin in exec process API
│
o #18085 [1/6] Add MCP server environment config
│
o main
```
---------
Co-authored-by: Codex <noreply@openai.com>
## Summary
- honor `_meta["codex/imageDetail"] == "original"` on MCP image content
and map it to `detail: "original"` where supported
- strip that detail back out when the active model does not support
original-detail image inputs
- update code-mode `image(...)` to accept individual MCP image blocks
- teach `js_repl` / `codex.emitImage(...)` to preserve the same hint
from raw MCP image outputs
- document the new `_meta` contract and add generic RMCP-backed coverage
across protocol, core, code-mode, and js_repl paths
## Changes
Allows MCPs to opt in to receiving sandbox config info through `_meta`
on model-initiated tool calls. This lets MCPs adhere to the thread's
sandbox if they choose to.
## Details
- Adds the `codex/sandbox-state-meta` experimental MCP capability.
- Tracks whether each MCP server advertises that capability.
- When a server opts in, `codex-core` injects the current `SandboxState`
into model-initiated MCP tool-call request `_meta`.
## Verification
- added an integration test for the capability
## Why
For more advanced MCP usage, we want the model to be able to emit
parallel MCP tool calls and have Codex execute eligible ones
concurrently, instead of forcing all MCP calls through the serial block.
The main design choice was where to thread the config. I made this
server-level because parallel safety depends on the MCP server
implementation. Codex reads the flag from `mcp_servers`, threads the
opted-in server names into `ToolRouter`, and checks the parsed
`ToolPayload::Mcp { server, .. }` at execution time. That avoids relying
on model-visible tool names, which can be incomplete in
deferred/search-tool paths or ambiguous for similarly named
servers/tools.
## What was added
Added `supports_parallel_tool_calls` for MCP servers.
Before:
```toml
[mcp_servers.docs]
command = "docs-server"
```
After:
```toml
[mcp_servers.docs]
command = "docs-server"
supports_parallel_tool_calls = true
```
MCP calls remain serial by default. Only tools from opted-in servers are
eligible to run in parallel. Docs also now warn to enable this only when
the server’s tools are safe to run concurrently, especially around
shared state or read/write races.
## Testing
Tested with a local stdio MCP server exposing real delay tools. The
model/Responses side was mocked only to deterministically emit two MCP
calls in the same turn.
Each test called `query_with_delay` and `query_with_delay_2` with `{
"seconds": 25 }`.
| Build/config | Observed | Wall time |
| --- | --- | --- |
| main with flag enabled | serial | `58.79s` |
| PR with flag enabled | parallel | `31.73s` |
| PR without flag | serial | `56.70s` |
PR with flag enabled showed both tools start before either completed;
main and PR-without-flag completed the first delay before starting the
second.
Also added an integration test.
Additional checks:
- `cargo test -p codex-tools` passed
- `cargo test -p codex-core
mcp_parallel_support_uses_exact_payload_server` passed
- `git diff --check` passed
- Changed `requires_mcp_tool_approval` to apply MCP spec defaults when
annotations are missing.
- Unannotated tools now default to:
- `readOnlyHint = false`
- `destructiveHint = true`
- `openWorldHint = true`
- This means unannotated MCP tools now go through approval/ARC
monitoring instead of silently bypassing it.
- Explicitly read-only tools still skip approval unless they are also
explicitly marked destructive.
**Previous behavior**
Failed open for missing annotations, which was unsafe for custom MCP
tools that omitted or forgot annotations.
---------
Co-authored-by: colby-oai <228809017+colby-oai@users.noreply.github.com>
## What?
- Render an MCP image output cell whenever a decodable image block
exists in `CallToolResult.content` (including text-before-image or
malformed image before valid image).
## Why?
- Tool results that include caption text before the image currently drop
the image output cell.
- A malformed image block can also suppress later valid image output.
## How?
- Iterate `content` and return the first successfully decoded image
instead of only checking the first block.
- Add unit tests that cover text-before-image ordering and
invalid-image-before-valid.
## Before
```rust
let image = match result {
Ok(mcp_types::CallToolResult { content, .. }) => {
if let Some(mcp_types::ContentBlock::ImageContent(image)) = content.first() {
// decode image (fails -> None)
} else {
None
}
}
_ => None,
}?;
```
## After
```rust
let image = result
.as_ref()
.ok()?
.content
.iter()
.find_map(decode_mcp_image)?;
```
## Risk / Impact
- Low: only affects image cell creation for MCP tool results; no change
for non-image outputs.
## Tests
- [x] `just fmt`
- [x] `cargo test -p codex-tui`
- [x] Rerun after branch update (2026-01-27): `just fmt`, `cargo test -p
codex-tui`
Manual testing
# Manual testing: MCP image tool result rendering (Codex TUI)
# Build the rmcp stdio test server binary:
cd codex-rs
cargo build -p codex-rmcp-client --bin test_stdio_server
# Register the server as an MCP server (absolute path to the built binary):
codex mcp add mcpimg -- /Users/joshka/code/codex-pr-review/codex-rs/target/debug/test_stdio_server
# Then in Codex TUI, ask it to call:
- mcpimg.image_scenario({"scenario":"image_only"})
- mcpimg.image_scenario({"scenario":"text_then_image","caption":"Here is the image:"})
- mcpimg.image_scenario({"scenario":"invalid_base64_then_image"})
- mcpimg.image_scenario({"scenario":"invalid_image_bytes_then_image"})
- mcpimg.image_scenario({"scenario":"multiple_valid_images"})
- mcpimg.image_scenario({"scenario":"image_then_text","caption":"Here is the image:"})
- mcpimg.image_scenario({"scenario":"text_only","caption":"Here is the image:"})
# Expected:
# - You should see an extra history cell: "tool result (image output)" when the
# tool result contains at least one decodable image block (even if earlier
# blocks are text or invalid images).
Fixes#9814
---------
Co-authored-by: Josh McKinney <joshka@openai.com>
It's pretty amazing we have gotten here without the ability for the
model to see image content from MCP tool calls.
This PR builds off of 4391 and fixes#4819. I would like @KKcorps to get
adequete credit here but I also want to get this fix in ASAP so I gave
him a week to update it and haven't gotten a response so I'm going to
take it across the finish line.
This test highlights how absured the current situation is. I asked the
model to read this image using the Chrome MCP
<img width="2378" height="674" alt="image"
src="https://github.com/user-attachments/assets/9ef52608-72a2-4423-9f5e-7ae36b2b56e0"
/>
After this change, it correctly outputs:
> Captured the page: image dhows a dark terminal-style UI labeled
`OpenAI Codex (v0.0.0)` with prompt `model: gpt-5-codex medium` and
working directory `/codex/codex-rs`
(and more)
Before this change, it said:
> Took the full-page screenshot you asked for. It shows a long,
horizontally repeating pattern of stylized people in orange, light-blue,
and mustard clothing, holding hands in alternating poses against a white
background. No text or other graphics-just rows of flat illustration
stretching off to the right.
Without this change, the Figma, Playwright, Chrome, and other visual MCP
servers are pretty much entirely useless.
I tested this change with the openai respones api as well as a third
party completions api
This PR adds support for streamable HTTP MCP servers when the
`experimental_use_rmcp_client` is enabled.
To set one up, simply add a new mcp server config with the url:
```
[mcp_servers.figma]
url = "http://127.0.0.1:3845/mcp"
```
It also supports an optional `bearer_token` which will be provided in an
authorization header. The full oauth flow is not supported yet.
The config parsing will throw if it detects that the user mixed and
matched config fields (like command + bearer token or url + env).
The best way to review it is to review `core/src` and then
`rmcp-client/src/rmcp_client.rs` first. The rest is tests and
propagating the `Transport` struct around the codebase.
Example with the Figma MCP:
<img width="5084" height="1614" alt="CleanShot 2025-09-26 at 13 35 40"
src="https://github.com/user-attachments/assets/eaf2771e-df3e-4300-816b-184d7dec5a28"
/>