Agent Blueprint MCP Server

agentblueprint

CLI and MCP server for Agent Blueprint. Create structured business profiles, generate pipeline artifacts, explore blueprint data, sync implementation state, and download full Agent Skills directories. Vendor-agnostic output works with ServiceNow, Salesforce, OpenClaw, or any platform.

Agent Blueprint is delivered through hands-on engagements; API tokens are issued to client teams. If your organization works with Agent Blueprint, this package connects your coding agents to your blueprint registry.

Not a client? You don't need this package. Agent Blueprint also has free tools that need no account or token:

• Draw your agent: have your coding agent draw a one-page blueprint of itself (what it can read, what it can change, who has to say yes) at agentblueprint.ai/personal. No sign-up.
• Remote MCP server: https://agentblueprint.ai/mcp (Streamable HTTP). No sign-in, read-only, keeps nothing. Tools: list_benchmark_jobs, estimate_agent_cost, check_agent_one_page, render_agent_map. Instructions for agents are at agentblueprint.ai/agents.
• One-page agent template: agentblueprint.ai/agent-one-page-template.
• Agent cost benchmark: measured cost and tokens for ten everyday jobs on Claude Code and Codex at agentblueprint.ai/agent-cost-benchmark.

CLI Quick Start

# Install globally
npm install -g agentblueprint

# Store your API token (one-time)
agentblueprint login

# List blueprints
agentblueprint list

# Get a blueprint summary (JSON to stdout)
agentblueprint get blueprint <id>

# Get other artifacts
agentblueprint get business-case <id>
agentblueprint get use-case <id>
agentblueprint get implementation-plan <id>
agentblueprint get implementation-spec <id>
agentblueprint get business-profile

# Download as Agent Skills directory
agentblueprint download <id>

# Partner cross-org access
agentblueprint list --org <customer-org-id>
agentblueprint get blueprint <id> --org <customer-org-id>

Or run without installing via npx:

npx agentblueprint list --token <your-api-key>
npx agentblueprint get blueprint <id> --token <your-api-key>

MCP Server Setup

The same binary auto-detects MCP mode when stdin is piped (non-interactive). No separate command needed.

Add to your Claude Code MCP config (.claude/settings.json or project settings):

{
  "mcpServers": {
    "agent-blueprint": {
      "command": "npx",
      "args": ["-y", "agentblueprint", "serve"],
      "env": {
        "AGENT_BLUEPRINT_API_KEY": "<your-api-key>"
      }
    }
  }
}

You can also start the MCP server explicitly with agentblueprint serve.

Get an API Key

API access comes with an Agent Blueprint engagement. If your organization is a client:

1. Go to Agent Blueprint Settings > API Tokens
2. Click "Create Token"
3. Copy the token (shown once)

Not a client yet? Get in touch at agentblueprint.ai.

Supported Headless Workflow

Current scope is intentionally narrow:

existing PAT + existing org -> create_business_profile -> generate_use_cases -> generate_blueprint or trigger_full_pipeline -> get_generation_status -> download_blueprint

This release does not handle organization creation, OAuth bootstrap, or free-form context extraction into the business profile.

Download Blueprint as Agent Skills

Download a blueprint as a local Agent Skills directory that any coding agent can read from the filesystem. This is the recommended way to work with blueprints.

# Using the CLI (after `agentblueprint login`)
agentblueprint download <id>
agentblueprint download <id> --dir ./my-skills

# Or via npx
npx agentblueprint download --token <key> --blueprint <id>

This creates an Agent Skills directory structure:

.agent-blueprint/<blueprint-slug>/
├── SKILL.md                              # Overview + frontmatter (auto-discovered by agents)
├── references/
│   ├── business-context.md               # Use case, pain points, transformation story
│   ├── agent-specifications.md           # Full agent specs with tools, guardrails, metrics
│   ├── financial-case.md                 # ROI, cost breakdown, sensitivity, 5-year projection
│   ├── implementation-roadmap.md         # Epics, stories, timeline, roles, dependencies
│   ├── architecture-decisions.md         # Platform, pattern, integration gaps, feasibility
│   └── guardrails-and-governance.md      # Risks, mitigation, per-agent guardrails
└── scripts/
    └── validate-spec.sh                  # Structure completeness checker

The Agent Skills standard is supported by Claude Code, Codex, Cursor, GitHub Copilot, Windsurf, and 18+ other coding agents. SKILL.md loads automatically at activation (~100 tokens), reference files load on demand.

Available Tools

The generate_blueprint and trigger_full_pipeline tools are asynchronous. Use get_generation_status to poll until the run completes, then call download_blueprint with the resolved blueprintId.

The get_blueprint, get_business_case, and get_implementation_plan tools return concise summaries optimized for agent context windows. For full details (agent specs, financial projections, user stories), use download_blueprint to get the complete Agent Skills directory.

Available Resources

Authentication

Three ways to provide your API token (checked in this order):

1. --token <key> flag on any command
2. AGENT_BLUEPRINT_API_KEY environment variable
3. agentblueprint login (saved to ~/.config/agentblueprint/config.json)

Environment Variables