Getting started
{
"mcpServers": {
"t3mcp": {
"command": "npx",
"args": [
"-y",
"@wyrd-company/t3code-mcp",
"--profile",
"sample"
]
}
}
} Getting started with t3mcp
@wyrd-company/t3code-mcp supplies stdio tools for scheduler endpoints, webhook
diagnostics, execution configuration, structured thread inspection, provider
status, historical usage and Stop/queue recovery. Use it beside the built-in T3 Code MCP server.
Stock tools handle ordinary schedule CRUD, run-now, launches, global thread
search, questions, model discovery and queued-message editing.
Use Node >=22.16 and the same saved profiles as t3c. Configure your MCP host
to launch the npm package with npx:
{
"command": "npx",
"args": ["-y", "@wyrd-company/t3code-mcp", "--profile", "sample"]
}
You can also install the executable with npm install -g @wyrd-company/t3code-mcp and use t3code-mcp --profile sample.
The install selector provides configurations for Claude Code, Claude Desktop, Codex, Cursor, OpenCode, Pi and Antigravity, plus generic JSON. Select your host and merge its configuration into the host’s MCP configuration file:
| Host | Configuration file |
|---|---|
| Claude Code | .mcp.json in the project |
| Claude Desktop | claude_desktop_config.json (Settings, Developer, Edit Config) |
| Codex | ~/.codex/config.toml |
| Cursor | .cursor/mcp.json in the project, or ~/.cursor/mcp.json |
| OpenCode | opencode.json in the project |
| Pi | ~/.pi/agent/mcp.json, or .pi/mcp.json in the project |
| Antigravity | ~/.gemini/config/mcp_config.json, or .agents/mcp_config.json in the project |
Replace sample with your saved profile name. All entries launch the same stdio
server with npx -y @wyrd-company/t3code-mcp --profile sample; Codex uses TOML
and OpenCode uses a local command array.
Pair the profile with t3c auth pair first. --profile selects an exact saved
profile and ignores environment overrides. Alternatively, --server and
--credentials use the same resolution as t3c. A named profile conflicts with
direct flags. Without a named profile, T3_SERVER_URL, T3_TOKEN and the current
t3c profile apply. The process connects when a tool first needs the server.
t3code-mcp tools lists the tool names without connecting.
t3mcp uses the direct route: pair its profile with the server’s own address. If people share the server through t3c-proxy, do not pair t3mcp with a person’s hostname, because a hostname is for the messages of one person.
| Tools | Result |
|---|---|
scheduled_task_endpoint | Relative token-bearing path, direct URL, optional Connect URL and signing-secret presence |
scheduled_task_deliveries, scheduled_task_delivery | Delivery outcomes, errors and request details |
rotate_scheduled_task_webhook | New endpoint; the previous URL stops working |
scheduled_task_configuration, configure_scheduled_task | Full task configuration/run diagnostics; execution-only changes |
thread_status, thread_approvals | Run/queue/background-work state and pending approvals for human review |
thread_usage, thread_diff, thread_search_content | Usage ownership, checkpoint diff or within-thread search |
thread_family, worktree_threads | A thread’s whole family, or the active threads on a worktree path |
thread_tag, tagged_threads | Add or remove a thread tag, or list the threads that carry a tag |
thread_terminals, terminal_output | A thread’s terminals, or the recent plain-text output of one |
server_traces, server_processes | The server’s trace summary, or its backend descendant processes |
server_host_resources, server_resource_history | Host CPU and memory, or the server’s resource use over a window |
server_storage_cleanup_report | The latest storage cleanup report |
local_servers, stop_local_server | Local servers and their owning threads, or stop one that a thread’s agent started |
thread_stop, thread_queue_resume, thread_queue_start | Native Stop, queue release or selected queued-message start |
thread_force_stop | Force stop of an agent that ignored Stop |
provider_status, usage_summary | Current provider state/limits or historical usage buckets |
Every tool returns matching structured data and JSON text. Failures set
isError and return a stable code with fixed text. Protocol cancellation reaches
the library operation. The server adds no per-tool deadline.
Scheduler endpoints, deliveries and mutations require orchestration:operate;
usage summaries require diagnostics:read. The selected credential controls
access. The server withholds webhook endpoints from read-only credentials. A
direct URL is useful without Connect, but it is reachable only where the selected
server is reachable. Delivery bodies, headers and rendered prompts are untrusted
request data. Signing secrets remain write-only.
configure_scheduled_task accepts {id, configuration}. Its configuration
contains projectId, nullable threadId, workspaceStrategy, modelSelection,
runtimeMode and interactionMode. Omitted settings stay unchanged; null threadId
starts a fresh thread per run. The update reads then writes and can overwrite a
concurrent edit. Use stock tools for trigger, prompt, title and enablement changes.
Changing the project or thread validates that the resulting bound thread belongs
to the resulting project before saving.
thread_diff accepts an optional range with fromTurnCount and toTurnCount; zero
means the thread start. Within-thread search returns a count and a selected match
with native navigation positions. Usage results keep source and pricing metadata.
Missing usage is not zero; session counts overlap across time buckets.
thread_stop holds queued messages and stops active/background work even when
no run is active. thread_queue_resume releases the hold. thread_queue_start
selects a queued plain user message by runId and releases the queue hold, so
other queued messages can follow. Get ids from thread_status or the stock
queue tools. These tools require orchestration:operate and return dispatch
acknowledgements; inspect status for subsequent state. Stock queue editing,
cancellation, reordering and promotion remain in the built-in MCP server.
thread_force_stop is for an agent that ignored thread_stop. It ends the turn
and kills or releases the agent process, and returns the outcome and the threads
that share that process. With runId, it fails with code superseded when
another run is the latest.
Build from source
From a repository checkout, run task install and task build. Configure the
host to run node with the absolute path to
packages/t3code-mcp/dist/cli.js, followed by --profile sample. Source builds
use the same connection settings and tool inventory as the npm package.