This PR adds the ability to favorite models for external agents—writing
to the settings in the `agent_servers` key—as well as a handful of other
improvements:
- Make the cycling keybinding `alt-enter` work for the inline assistant
as well as previous user messages
- Better organized the keybinding files removing some outdated
agent-related keybinding definitions
- Renamed the inline assistant key context to "InlineAssistant" as
"PromptEditor" is old and confusing
- Made the keybindings to rate an inline assistant response visible in
the thumbs up/down button's tooltip
- Created a unified component for the model selector tooltip given we
had 3 different places creating the same element
- Make the "Cycle Favorited Models" row in the tooltip visible only if
there is more than one favorite models
Release Notes:
- agent: External agents also now support the favoriting model feature,
which comes with a handy keybinding to cycle through the favorite list.
### Summary:
- Ensure the external agents with ACP servers start via non-interactive
shells to prevent shell startup noise from corrupting JSON-RPC.
- Apply the same tweak to MCP stdio transports so remote context servers
aren’t affected by prompts or greetings.
### Description:
Switch both ACP and MCP stdio launch paths to call
`ShellBuilder::non_interactive()` before building the command. This
removes `-i` on POSIX shells, suppressing prompt/title sequences that
previously prefixed the first JSON line and caused `serde_json` parse
failures. No functional regressions are expected: both code paths only
need a shell for Windows/npm script compatibility, not for
interactivity.
Release Notes:
- Fixed external agents that hung on “Loading…” when shell startup
output broke JSON-RPC initialization.
We were defining these in multiple places and also weren't leveraging
the ids the agents were already providing.
This should make sure we use them consistently and avoid issues in the
future.
Release Notes:
- N/A
Uses the latest version of the SDK + schema crate. A bit painful because
we needed to move to `#[non_exhaustive]` on all of these structs/enums,
but will be much easier going forward.
Also, since we depend on unstable features, I am pinning the version so
we don't accidentally introduce compilation errors from other update
cycles.
Release Notes:
- N/A
This allows you to specify default_model and default_mode for ACP
extensions, e.g.
```
"auggie": {
"default_model": "gpt-5",
"default_mode": "default",
"type": "extension"
},
```
Release Notes:
- Added support for specifying settings for ACP extensions
(`default_mode`, `default_model`)
This PR builds on top of the `default_mode` feature where it was
possible to set an external agent mode as the default if you held a
modifier while clicking on the desired option. Now, if you want to have,
for example, Haiku as your default Claude Code model, you can do that.
This feature adds parity between external agents and Zed's built-in one,
which already supported this feature for a little while.
Note: This still doesn't work with external agents installed from
extensions. At the moment, this is limited to Claude Code, Codex, and
Gemini—the ones we include out of the box.
Release Notes:
- agent: Added the ability to set a model as the default for a given
built-in external agent (Claude Code, Codex CLI, or Gemini CLI).
### What this solves
This PR adds support for HTTP and SSE (Server-Sent Events) transports to
Zed's context server implementation, enabling communication with remote
MCP servers. Currently, Zed only supports local MCP servers via stdio
transport. This limitation prevents users from:
- Connecting to cloud-hosted MCP servers
- Using MCP servers running in containers or on remote machines
- Leveraging MCP servers that are designed to work over HTTP/SSE
### Why it's important
The MCP (Model Context Protocol) specification includes HTTP/SSE as
standard transport options, and many MCP server implementations are
being built with these transports in mind. Without this support, Zed
users are limited to a subset of the MCP ecosystem. This is particularly
important for:
- Enterprise users who need to connect to centralized MCP services
- Developers working with MCP servers that require network isolation
- Users wanting to leverage cloud-based context providers (e.g.,
knowledge bases, API integrations)
### Implementation approach
The implementation follows Zed's existing architectural patterns:
- **Transports**: Added `HttpTransport` and `SseTransport` to the
`context_server` crate, built on top of the existing `http_client` crate
- **Async handling**: Uses `gpui::spawn` for network operations instead
of introducing a new Tokio runtime
- **Settings**: Extended `ContextServerSettings` enum with a `Remote`
variant to support URL-based configuration
- **UI**: Updated the agent configuration UI with an "Add Remote Server"
option and dedicated modal for remote server management
### Changes included
- [x] HTTP transport implementation with request/response handling
- [x] SSE transport for server-sent events streaming
- [x] `build_transport` function to construct appropriate transport
based on URL scheme
- [x] Settings system updates to support remote server configuration
- [x] UI updates for adding/editing remote servers
- [x] Unit tests using `FakeHttpClient` for both transports
- [x] Integration tests (WIP)
- [x] Documentation updates (WIP)
### Testing
- Unit tests for both `HttpTransport` and `SseTransport` using mocked
HTTP client
- Manual testing with example MCP servers over HTTP/SSE
- Settings validation and UI interaction testing
### Screenshots/Recordings
[TODO: Add screenshots of the new "Add Remote Server" UI and
configuration modal]
### Example configuration
Users can now configure remote MCP servers in their `settings.json`:
```json
{
"context_servers": {
"my-remote-server": {
"enabled": true,
"url": "http://localhost:3000/mcp"
}
}
}
```
### AI assistance disclosure
I used AI to help with:
- Understanding the MCP protocol specification and how HTTP/SSE
transports should work
- Reviewing Zed's existing patterns for async operations and suggesting
consistent approaches
- Generating boilerplate for test cases
- Debugging SSE streaming issues
All code has been manually reviewed, tested, and adapted to fit Zed's
architecture. The core logic, architectural decisions, and integration
with Zed's systems were done with human understanding of the codebase.
AI was primarily used as a reference tool and for getting unstuck on
specific technical issues.
Release notes:
* You can now configure MCP Servers that connect over HTTP in your
settings file. These are not yet available in the extensions API.
```
{
"context_servers": {
"my-remote-server": {
"enabled": true,
"url": "http://localhost:3000/mcp"
}
}
}
```
---------
Co-authored-by: Conrad Irwin <conrad.irwin@gmail.com>
As discussed with @benbrandt and @mikayla-maki:
* We now tell ACP clients we support the nonstandard `terminal-auth`
`_meta` field for terminal-based authentication
* In the future, we anticipate ACP itself supporting *some* form of
terminal-based authentication, but that hasn't been designed yet or gone
through the RFD process
* For now, this unblocks terminal-based auth
Release Notes:
- Added experimental terminal-based authentication to ACP support
Updates to acp crate 0.7, which allows us to send information about the
client to the Agent.
In the future, we can also use the AgentInfo on the response for
internal metrics.
Release Notes:
- N/A
We've been considering removing workspace-hack for a couple reasons:
- Lukas ran into a situation where its build script seemed to be causing
spurious rebuilds. This seems more likely to be a cargo bug than an
issue with workspace-hack itself (given that it has an empty build
script), but we don't necessarily want to take the time to hunt that
down right now.
- Marshall mentioned hakari interacts poorly with automated crate
updates (in our case provided by rennovate) because you'd need to have
`cargo hakari generate && cargo hakari manage-deps` after their changes
and we prefer to not have actions that make commits.
Currently removing workspace-hack causes our workspace to grow from
~1700 to ~2000 crates being built (depending on platform), which is
mainly a problem when you're building the whole workspace or running
tests across the the normal and remote binaries (which is where
feature-unification nets us the most sharing). It doesn't impact
incremental times noticeably when you're just iterating on `-p zed`, and
we'll hopefully get these savings back in the future when
rust-lang/cargo#14774 (which re-implements the functionality of hakari)
is finished.
Release Notes:
- N/A
This applies the same change as #39466 to the terminal codepath for
external agents.
Release Notes:
- N/A
Co-authored-by: Max Brunsfeld <maxbrunsfeld@gmail.com>
Closes#39614
The `ShellKind` struct is built on Windows' side, meaning that when
connecting to remotes, we fall back to PowerShell construction, even if
the shell program we are spawning is a unix program.
This broke tasks creation since we are using the shell kind to construct
args:
d04ac864b8/crates/project/src/terminals.rs (L149)
In normal terminals this only affected activation scripts (only place
where shell kind is used)
I don't have a Windows machine to test it, so I would appreciate any
help with testing!
Release Notes:
- Fixed an issue where tasks could not be executed in Windows WSL
---------
Signed-off-by: Marco Mihai Condrache <52580954+marcocondrache@users.noreply.github.com>
If we get a `cwd` from ACP (because e.g. `codex-acp` is driving the
terminal rather than our own PTY) then use that to display the `cwd` of
the terminal process.
Release Notes:
- N/A
This only affects `codex-acp` for now.
Not using the PTY in display-only terminals means they don't display the
login prompt (or spurious `%`s) at the end of terminal output
renderings.
Release Notes:
- N/A
Codex needs (and future projects are anticipated to need as well) a
concept of display-only terminals. This refactors terminals to decouple
the PTY part from the display part, so that we can render terminal
changes based on a series of events - regardless of whether they're
being driven from a PTY inside Zed or from an outside source (e.g.
`codex-acp`).
Release Notes:
- N/A
Since we might run MCP servers locally for an agent, we don't want to
use the proxy for those.
We set this if the user has set a proxy, but not a custom NO_PROXY env
var.
Closes#38839
Release Notes:
- acp: Don't run local mcp servers through proxy, if set
Co-Authored-By: Ben K <ben@zed.dev>
Co-Authored-By: Anthony <anthony@zed.dev>
Co-Authored-By: Mikayla <mikayla@zed.dev>
Release Notes:
- settings: Major internal changes to settings. The primary user-facing
effect is that some settings which did not make sense in project
settings files are no-longer read from there. (For example the inline
blame settings)
---------
Co-authored-by: Ben Kunkle <ben@zed.dev>
Co-authored-by: Mikayla Maki <mikayla.c.maki@gmail.com>
Co-authored-by: Anthony <anthony@zed.dev>
- Use ProxySettings::proxy_url to read from settings or env
- Export HTTP(S)_PROXY and NO_PROXY for agent CLIs
- Add read_no_proxy_from_env and move parsing from main
Closes https://github.com/zed-industries/claude-code-acp/issues/46
Release Notes:
- acp: Pass proxy settings through to all ACP agents
Three motivations for this:
* Changing provider URL could cause credentials for the prior URL to be
sent to the new URL.
* The UI is in a misleading state after URL change - it shows a
configured API key, but on restart it will show no API key.
* #34110 will add support for both URL and key configuration for Ollama.
This is the first provider to have UI for setting the URL, and this
makes these issues show up more directly as odd UI interactions.
#37610 implemented something similar for the OpenAI and OpenAI
compatible providers. This extracts out some shared code, uses it in all
relevant providers, and adds more safety around key use.
I haven't tested all providers, but the per-provider changes were pretty
mechanical, so hopefully work properly.
Release Notes:
- Fixed handling of changes to LLM provider URL in settings to also load
the associated API key.
It appears that in macOS, the `AcpConnection._wait_task` doesn't always
get dropped when quitting the app. In these cases, the subprocess would
be kept alive because we move the `child` into it.
Instead, we will now explicitly kill it when `AcpConnection` is dropped.
It's ok to do this because when the connection is dropped, the thread is
also dropped, so there's no need to report the exit status to it.
Closes#37741
Release Notes:
- Claude Code: Fix subprocess leak on app quit
Closes#37487
Proxy settings are now taken from the Zed configuration and passed to
Gemini via the "--proxy" flag.
Release Notes:
- acp: Gemini ACP server now uses proxy settings from Zed configuration.
This PR adds support for using external agents in SSH projects via ACP,
including automatic installation of Gemini CLI and Claude Code,
authentication with API keys (for Gemini) and CLI login, and custom
agents from user configuration.
Co-authored-by: maan2003 <manmeetmann2003@gmail.com>
Release Notes:
- agent: Gemini CLI, Claude Code, and custom external agents can now be
used in SSH projects.
---------
Co-authored-by: maan2003 <manmeetmann2003@gmail.com>
Closes#37469
Release Notes:
- agent: The project shell environment is now passed to external agent
processes.
Co-authored-by: Richard Feldman <oss@rtfeldman.com>
Co-authored-by: Nia Espera <nia-e@haecceity.cc>
This PR separates out the associated constant `KEY` from the `Settings`
trait into a new trait `SettingsKey`. This allows for the key trait to
be derived using attributes to specify the path so that the new
`SettingsUi` derive macro can use the same attributes to determine top
level settings paths thereby removing the need to duplicate the path in
both `Settings::KEY` and `#[settings_ui(path = "...")]`
Co-authored-by: Ben Kunkle <ben@zed.dev>
Release Notes:
- N/A
---------
Co-authored-by: Ben Kunkle <ben@zed.dev>
Since Claude Code has it's own preferred method of grabbing API keys, we
don't want to reuse this one.
Release Notes:
- acp: Don't share Anthropic API key from the Anthropic provider to
allow default Claude Code login options
---------
Co-authored-by: Agus Zubiaga <agus@zed.dev>
