A workflow CLI your agents can use.
Claude Code, Codex and ChatGPT turn what you ask into workflows that run on their own.
Claude Code · ~/sales
› Every weekday at 08:00, score new leads and post the hot ones to #sales.
$ tutti workflows create --file morning-leads.json
id wf_8f2k1
name Morning leads
…
$ tutti workflows show "Morning leads"
workflow: Morning leads
◆ Weekdays · Schedule
└─→ ● New leads · records/list-people
└─→ ● For each lead · system/loop
├─→ ● Score the lead · anthropic/generate-text
├─→ ● Hot leads only · system/condition
│ └─→ ● Post to #sales · slack/send-message
│ └─→ ● Create a customer · stripe/create-customer
│ └─→ ● Send the digest · google-gmail/send-gmail
└─→ ● Add to nurture list · records/upsert-person
└─→ ↩ Send the digest
Morning leads
Weekdays at 08:00
Weekdays 08:00Schedule
+
New leadsLibrary
+
For each lead
Score the leadClaude
+
Hot leads onlyCondition
+
Post to #salesSlack
+
Create a customerStripe
Add to nurture listLibrary
+
Send the digestGmail
Claude Code · ~/sales
› Every weekday at 08:00, score new leads and post the hot ones to #sales.
$ tutti workflows create --file morning-leads.json
id wf_8f2k1
name Morning leads
…
$ tutti run "Morning leads"
execution id exec_3kq9
workflow id wf_8f2k1
workflow name Morning leads
status success
duration 4612
✱ Done. Morning leads runs every weekday at 08:00, without me.
Your agent does a job once. A workflow does it every day.
Your agent with Tutti
MondayYou ask it once
TuesdayRan at 08:00 · Done in 4.6 s
WednesdayRan at 08:00 · Done in 4.2 s
ThursdayRan at 08:00 · Done in 4.9 s
FridayRan at 08:00 · Done in 4.4 s
Your agent alone
MondayYou ask it
TuesdayYou ask again
WednesdayYou ask again
ThursdayYou forget
FridayYou ask again
Your agent with Tutti
MondayYou ask it once
TuesdayRan at 08:00 · Done in 4.6 s
WednesdayRan at 08:00 · Done in 4.2 s
ThursdayRan at 08:00 · Done in 4.9 s
FridayRan at 08:00 · Done in 4.4 s
Two ways in: a CLI and an MCP server.
The CLIFor agents in a terminal, like Claude Code and Codex
The MCP serverFor agents you chat with, like Claude and ChatGPT. 121 tools.
The CLI
For agents that work in a terminal, like Claude Code and Codex.
npm i -g @darkmatter-ai/tutti
tutti workflows create --file wf.json
tutti run "Morning leads"
The MCP server
For agents you chat with, like Claude and ChatGPT.
create_workflow
execute_workflow
get_execution_logs
The CLI answers in JSON your agent can read.
Terminal
$ tutti run "Morning leads" --raw
{
"executionId": "exec_3kq9",
"workflowName": "Morning leads",
"status": "success",
"duration": "4612"
}
Terminal
$ tutti run "Morning leads" --raw
{ "executionId": "exec_3kq9", "workflowId": "wf_8f2k1", "workflowName": "Morning leads", "status": "success", "duration": "4612" }
Piped output is always JSON.Errors come back as JSON, with an exit code.Anything destructive needs --yes.
When a run fails, your agent can fix it.
Claude Code
$ tutti executions logs exec_7m2p \
--status error
{ "nodeName": "Post to #sales",
"error": "channel_not_found" }
$ tutti workflows node patch \
"Morning leads" post \
--set slackChannel=#sales-team
$ tutti run "Morning leads"
{ "status": "success", ... }
Claude Code
$ tutti executions logs exec_7m2p --status error
{ "nodeName": "Post to #sales", "status": "error", "error": "Slack replied channel_not_found" }
$ tutti workflows node patch "Morning leads" post --set 'slackChannel=#sales-team'
{ "id": "post", "label": "Post to #sales", ... }
$ tutti run "Morning leads"
{ "status": "success", "duration": "4418", ... }
Anything you can do in Tutti, your agent can do too.
Each tool on the MCP server has a matching CLI command.
Records18
Team database13
Workflow9
Execution8
Notes and activity8
13 more groups65
Records: 18
Team database: 13
Workflow: 9
Execution: 8
Notes and activity: 8
Pipelines: 7
Reference: 7
Folders: 6
Integration: 6
Email templates: 6
Capture: 6
Connectors: 5
Custom columns: 5
Next steps: 5
Schedules: 4
Browser sessions: 4
WhatsApp: 3
Actions: 1
In all: 121
