feat: support mcp_servers in config.toml (#829)

This adds initial support for MCP servers in the style of Claude Desktop
and Cursor. Note this PR is the bare minimum to get things working end
to end: all configured MCP servers are launched every time Codex is run,
there is no recovery for MCP servers that crash, etc.

(Also, I took some shortcuts to change some fields of `Session` to be
`pub(crate)`, which also means there are circular deps between
`codex.rs` and `mcp_tool_call.rs`, but I will clean that up in a
subsequent PR.)

`codex-rs/README.md` is updated as part of this PR to explain how to use
this feature. There is a bit of plumbing to route the new settings from
`Config` to the business logic in `codex.rs`. The most significant
chunks for new code are in `mcp_connection_manager.rs` (which defines
the `McpConnectionManager` struct) and `mcp_tool_call.rs`, which is
responsible for tool calls.

This PR also introduces new `McpToolCallBegin` and `McpToolCallEnd`
event types to the protocol, but does not add any handlers for them.
(See https://github.com/openai/codex/pull/836 for initial usage.)

To test, I added the following to my `~/.codex/config.toml`:

```toml
# Local build of https://github.com/hideya/mcp-server-weather-js
[mcp_servers.weather]
command = "/Users/mbolin/code/mcp-server-weather-js/dist/index.js"
args = []
```

And then I ran the following:

```
codex-rs$ cargo run --bin codex exec 'what is the weather in san francisco'
[2025-05-06T22:40:05] Task started: 1
[2025-05-06T22:40:18] Agent message: Here’s the latest National Weather Service forecast for San Francisco (downtown, near 37.77° N, 122.42° W):

This Afternoon (Tue):
• Sunny, high near 69 °F
• West-southwest wind around 12 mph

Tonight:
• Partly cloudy, low around 52 °F
• SW wind 7–10 mph
...
```

Note that Codex itself is not able to make network calls, so it would
not normally be able to get live weather information like this. However,
the weather MCP is [currently] not run under the Codex sandbox, so it is
able to hit `api.weather.gov` and fetch current weather information.

---
[//]: # (BEGIN SAPLING FOOTER)
Stack created with [Sapling](https://sapling-scm.com). Best reviewed
with [ReviewStack](https://reviewstack.dev/openai/codex/pull/829).
* #836
* __->__ #829
This commit is contained in:
Michael Bolin
2025-05-06 15:47:59 -07:00
committed by GitHub
parent 49d040215a
commit 147a940449
11 changed files with 453 additions and 18 deletions

View File

@@ -31,6 +31,8 @@ use tracing::warn;
use crate::client::ModelClient;
use crate::client::Prompt;
use crate::client::ResponseEvent;
use crate::config::Config;
use crate::config::ConfigOverrides;
use crate::error::CodexErr;
use crate::error::Result as CodexResult;
use crate::exec::process_exec_tool_call;
@@ -38,6 +40,9 @@ use crate::exec::ExecParams;
use crate::exec::ExecToolCallOutput;
use crate::exec::SandboxType;
use crate::flags::OPENAI_STREAM_MAX_RETRIES;
use crate::mcp_connection_manager::try_parse_fully_qualified_tool_name;
use crate::mcp_connection_manager::McpConnectionManager;
use crate::mcp_tool_call::handle_mcp_tool_call;
use crate::models::ContentItem;
use crate::models::FunctionCallOutputPayload;
use crate::models::ResponseInputItem;
@@ -188,9 +193,9 @@ impl Recorder {
/// Context for an initialized model agent
///
/// A session has at most 1 running task at a time, and can be interrupted by user input.
struct Session {
pub(crate) struct Session {
client: ModelClient,
tx_event: Sender<Event>,
pub(crate) tx_event: Sender<Event>,
ctrl_c: Arc<Notify>,
/// The session's current working directory. All relative paths provided by
@@ -202,6 +207,9 @@ struct Session {
sandbox_policy: SandboxPolicy,
writable_roots: Mutex<Vec<PathBuf>>,
/// Manager for external MCP servers/tools.
pub(crate) mcp_connection_manager: McpConnectionManager,
/// External notifier command (will be passed as args to exec()). When
/// `None` this feature is disabled.
notify: Option<Vec<String>>,
@@ -433,7 +441,7 @@ impl State {
}
/// A series of Turns in response to user input.
struct AgentTask {
pub(crate) struct AgentTask {
sess: Arc<Session>,
sub_id: String,
handle: AbortHandle,
@@ -554,6 +562,26 @@ async fn submission_loop(
};
let writable_roots = Mutex::new(get_writable_roots(&cwd));
// Load config to initialize the MCP connection manager.
let config = match Config::load_with_overrides(ConfigOverrides::default()) {
Ok(cfg) => cfg,
Err(e) => {
error!("Failed to load config for MCP servers: {e:#}");
// Fall back to empty server map so the session can still proceed.
Config::load_default_config_for_test()
}
};
let mcp_connection_manager =
match McpConnectionManager::new(config.mcp_servers.clone()).await {
Ok(mgr) => mgr,
Err(e) => {
error!("Failed to create MCP connection manager: {e:#}");
McpConnectionManager::default()
}
};
sess = Some(Arc::new(Session {
client,
tx_event: tx_event.clone(),
@@ -563,6 +591,7 @@ async fn submission_loop(
sandbox_policy,
cwd,
writable_roots,
mcp_connection_manager,
notify,
state: Mutex::new(state),
}));
@@ -753,11 +782,14 @@ async fn run_turn(
} else {
None
};
let extra_tools = sess.mcp_connection_manager.list_all_tools();
let prompt = Prompt {
input,
prev_id,
instructions,
store,
extra_tools,
};
let mut retries = 0;
@@ -1141,13 +1173,20 @@ async fn handle_function_call(
}
}
_ => {
// Unknown function: reply with structured failure so the model can adapt.
ResponseInputItem::FunctionCallOutput {
call_id,
output: crate::models::FunctionCallOutputPayload {
content: format!("unsupported call: {}", name),
success: None,
},
match try_parse_fully_qualified_tool_name(&name) {
Some((server, tool_name)) => {
handle_mcp_tool_call(sess, &sub_id, call_id, server, tool_name, arguments).await
}
None => {
// Unknown function: reply with structured failure so the model can adapt.
ResponseInputItem::FunctionCallOutput {
call_id,
output: crate::models::FunctionCallOutputPayload {
content: format!("unsupported call: {}", name),
success: None,
},
}
}
}
}
}