This PR factors the tool definitions out of the `assistant` crate so
that they can be shared between `assistant` and `assistant2`.
`ToolWorkingSet` now lives in `assistant_tool`. The tool definitions
themselves live in `assistant_tools`, with the exception of the
`ContextServerTool`, which has been moved to the `context_server` crate.
As part of this refactoring I needed to extract the
`ContextServerSettings` to a separate `context_server_settings` crate so
that the `extension_host`—which is referenced by the `remote_server`—can
name the `ContextServerSettings` type without pulling in some undesired
dependencies.
Release Notes:
- N/A
This optimizes and fixes bugs in our logic for maintaining a set of
running context servers, based on the combination of the user's
`context_servers` settings and their installed extensions.
Release Notes:
- N/A
---------
Co-authored-by: Marshall <marshall@zed.dev>
Co-authored-by: Marshall Bowers <elliott.codes@gmail.com>
This PR reworks how the Assistant Panel references slash commands,
context servers, and tools.
Previously we were always reading them from the global registries, but
now we store individual collections on each Assistant Panel instance so
that there can be different ones registered for each project.
Release Notes:
- N/A
---------
Co-authored-by: Max <max@zed.dev>
Co-authored-by: Antonio <antonio@zed.dev>
Co-authored-by: Joseph <joseph@zed.dev>
Co-authored-by: Max Brunsfeld <maxbrunsfeld@gmail.com>
This PR updates the `SlashCommand` trait to use a streaming return type.
This change is just at the trait layer. The goal here is to decouple
changing the trait's API while preserving behavior on either side.
The `SlashCommandOutput` type now has two methods for converting two and
from a stream to use in cases where we're not yet doing streaming.
On the `SlashCommand` implementer side, the implements can call
`to_event_stream` to produce a stream of events based off the
`SlashCommandOutput`.
On the slash command consumer side we use
`SlashCommandOutput::from_event_stream` to convert a stream of events
back into a `SlashCommandOutput`.
The `/file` slash command has been updated to emit `SlashCommandEvent`s
directly in order for it to work properly.
Release Notes:
- N/A
---------
Co-authored-by: Max <max@zed.dev>
This PR adds a new `SlashCommandResult` type alias.
We're going to be changing what slash commands can return in order to
support streaming, so having this type alias in place will make that
switch a bit more neat.
Release Notes:
- N/A
We sadly have to change the underlying protocol once again. This will
likely be the last change to the core protocol without correctly
handling older versions. From here on out, we want to get better with
version handling. To do so, we introduce the notion of a string protocol
version to be explicit of when the underlying protocol last changed.
The change also changes the return values of prompts. For now we only
allow User messages from servers to match the current behaviour. We will
change this once #19222 lands which will allow slash commands to insert
user and assistant messages.
Release Notes:
- N/A
This PR does two things. It fixes some minor inconsistencies in the
protocol. This is mostly about handling JSON RPC notifications correctly
and skipping fields when set to None.
Second part is about improving the rendering of context server commands,
by passing on the description
of the command to the slash command UI and showing the name of the
argument as a CodeLabel.
Release Notes:
- N/A
Release Notes:
- Added a new `/delta` command to re-insert changed files that were
previously included in a context.
---------
Co-authored-by: Roy <roy@anthropic.com>
We accidentally do not accept prompts with an empty list of arguments,
as opposed to non given arguments list. We need to allow these. We also
not really supporting non required arguments, despite the protocol
describing it. This is a first iteration on fixing this as well.
Release Notes:
- N/A
Context servers might return CR characters, which are not acceptable in
Zed and cause ranges to be invalidated. We need to normalize them.
Closes#17109
Release Notes:
- context_servers: Fixed an issue where context servers returning a
carriage return character would result in a panic.
This PR adds support for completions via MCP. The protocol now supports
a new request type "completion/complete"
that can either complete a resource URI template (which we currently
don't use in Zed), or a prompt argument.
We use this to add autocompletion to our context server slash commands!
https://github.com/user-attachments/assets/08c9cf04-cbeb-49a7-903f-5049fb3b3d9f
Release Notes:
- context_servers: Added support for argument completions for context
server prompts. These show up as regular completions to slash commands.
Server can now include an optional description in a `prompts/get`
response. Zed will displayed the description as label of the slash
command.
Release Notes:
- context_servers: Servers can provide an optional description in
`prompts/get` responses that is displayed as the slash command label.
This commit proposes the addition of "context serveres" and the
underlying protocol (model context protocol). Context servers allow
simple definition of slash commands in another language and running
local on the user machines. This aims to quickly prototype new commands,
and provide a way to add personal (or company wide) customizations to
the assistant panel, without having to maintain an extension. We can
use this to reuse our existing codebase, with authenticators, etc and
easily have it provide context into the assistant panel.
As such it occupies a different design space as extensions, which I
think are
more aimed towards long-term, well maintained pieces of code that can be
easily distributed.
It's implemented as a central crate for easy reusability across the
codebase
and to easily hook into the assistant panel at all points.
Design wise there are a few pieces:
1. client.rs: A simple JSON-RPC client talking over stdio to a spawned
server. This is
very close to how LSP work and likely there could be a combined client
down the line.
2. types.rs: Serialization and deserialization client for the underlying
model context protocol.
3. protocol.rs: Handling the session between client and server.
4. manager.rs: Manages settings and adding and deleting servers from a
central pool.
A server can be defined in the settings.json as:
```
"context_servers": [
{"id": "test", "executable": "python", "args": ["-m", "context_server"]
]
```
## Quick Example
A quick example of how a theoretical backend site can look like. With
roughly 100 lines
of code (nicely generated by Claude) and a bit of decorator magic (200
lines in total), one
can come up with a framework that makes it as easy as:
```python
@context_server.slash_command(name="rot13", description="Perform a rot13 transformation")
@context_server.argument(name="input", type=str, help="String to rot13")
async def rot13(input: str) -> str:
return ''.join(chr((ord(c) - 97 + 13) % 26 + 97) if c.isalpha() else c for c in echo.lower())
```
to define a new slash_command.
## Todo:
- Allow context servers to be defined in workspace settings.
- Allow passing env variables to context_servers
Release Notes:
- N/A
---------
Co-authored-by: Marshall Bowers <elliott.codes@gmail.com>
