/bland:* slash commands for a file-based pathway workspace, and Norm, the agent builder with a self-driving convergence loop.
One plugin, three layers:
- The hosted MCP server at
api.bland.ai/v1/mcpdoes the work: reads and writes against your account, the pathway compiler, call logs, analytics, and docs search. - Skills teach the agent how to use it well. They load on demand when the conversation matches: pathways, analytics, evals, call review, tools, personas, knowledge bases, automations, triage, debugging, the raw API, and setup.
- Norm and the
/bland:*commands turn a pathway into local files you edit like code, validate against the real compiler, simulate, and commit.
Prerequisites
- A Bland API key from app.bland.ai/settings/api-keys. Never paste it into the chat; each host stores it for you as described below.
- Node.js 18 or newer for the workspace commands and hooks. Claude Code ships with it.
Install
- Claude Code
- Claude Desktop
- Cursor
- Codex
Inside a Claude Code session:The install prompts for your API key and stores it in your OS keychain. Scripted installs pass it on the command line so it never enters the model context:Restart the session, then verify:
Upgrading from the Norm plugin
Version 2.0 renamed the plugin fromnorm to bland. Run /plugin uninstall norm@bland, then /plugin install bland@bland and enter your key when prompted. The new plugin does not migrate the old credentials, so the key must be configured once. Commands moved from /norm:* to /bland:*, and the domain commands (tools, analytics, evals, review, and the rest) became skills that load on demand: just ask.
The flow, start to production
- Build it.
/bland:norm build me a booking flow that collects name, callback number, and a time, then confirms it back. Norm designs the pathway, validates it with the real compiler, and saves it. - Give it tools. Ask for a tool that checks availability in your booking API. The tools skill builds it, test-runs it against the real endpoint, and attaches it to the node that needs it.
- Prove it works.
/bland:loop <pathway_id> --goal 'caller books an appointment and gets it confirmed back'. Norm simulates the customer, grades every outcome with evidence, fixes what fails, and repeats until the call passes. - Define what to measure. Ask for an outcome schema: booked yes/no, appointment time, reason if not booked. Verified on real calls before you trust it.
- Score every call. Ask for a scorecard for this pathway. Every production call auto-scores after it ends.
- Watch it. Ask for an ops dashboard. A live board in the Bland UI: volume, completion, outcomes, score trends.
- Iterate from reality. Ask which calls are behind a strange number, review the call to find the exact turn it broke, then
/bland:loop <pathway_id> --from-call <call_id>to fix the flow against that real call.
Edit an existing pathway
Norm turns a pathway into local files you can read and edit: prompts as Markdown, configuration as YAML.Command reference
Skills
Skills load automatically when the conversation matches their description, on every host that readsSKILL.md.
Safety
- Reads are free. Anything that mutates state, costs money, or dials a phone asks for confirmation first.
- Norm never invents numbers or verdicts. Every claim is backed by a query, a read-back, or a transcript quote.
- Your API key stays in the host’s configuration and flows only to the MCP connection.
Advanced: the setup skill explains how to point the plugin at a non-production API URL, for example a staging server. The MCP transport requires
https, so expose local servers through a tunnel.Related
Bland MCP Server
The hosted MCP endpoint behind the plugin.
Tool reference
Every tool the server exposes.
Norm
The in-app version of Norm in the Bland dashboard.
Conversational Pathways
The flow graph Norm builds and edits.
Docs for agents: llms.txt