Talk to your architecture through MCP.
Albumi ships a Model Context Protocol server. Connect any MCP-compatible AI agent - Claude, Cursor, or your own - and let it read the model, validate, audit, and propose changes through governed channels.
AI agents - Claude, Cursor, custom workflows - are increasingly part of how teams work. Feed the agent the material you already have - wiki pages, spreadsheets, onboarding docs, tribal knowledge - and let it draft a first version of the architecture catalog. Ask questions against the live model like "which applications carry PII data?" or "what's the average integration count per business capability?" and get an answer in seconds instead of rebuilding context from scratch.
Albumi exposes four MCP tools: pull, validate, audit, and push. The agent reads structured workspace data, drafts changes against the schema, and pushes proposals as Architecture Change Requests. Same governance applies: the push never lands as a direct write, and the audit trail treats AI as another author. AI integration is included on every plan.
# Install and authenticate npm install -g @albumi/cli albumi login # Add to Claude Code /plugin install albumi@albumi-cli /reload-plugins # Then talk to your agent: "Pull our workspace. Generate applications from this wiki page. Push as 'Q2 Onboarding'."
What the agent can actually do.
Pull the live workspace
pull_workspace returns typed architecture data: applications, integrations, data objects, capabilities, components, organizations, initiatives, and their relationships. The agent reasons about the real model shape instead of guessing from prose.
Validate before push
validate checks schema, required fields, enums, and referential integrity so the agent fixes broken references and invalid values before anything reaches review.
Run health checks
audit exposes the same Architecture Health checks as the UI, but as structured output the agent can reason about before deciding what to change.
Submit as reviewable change
push_workspace never writes directly. It opens an Architecture Change Request that goes through the same board, voting, and merge path as a human-created proposal.
Use the client you already like
Claude Code, Cursor, or any MCP-compatible client can connect through the same stdio transport and the same CLI. There is no special premium integration path.
Inherit real permissions
The agent acts with the permissions of the account behind albumi login - no hidden superuser path, no special write bypass, nothing more than the user could do in the UI.
Where teams use it first.
Bootstrap the model from documents
Point the agent at wiki pages, onboarding docs, spreadsheets, or architecture notes and let it prepare the first pass of applications, integrations, and data objects for review.
Fill in metadata at scale
Use the agent for repetitive work: descriptions, criticality, lifecycle cleanup, and other bulk edits that are painful in forms. Reviewers catch what the agent cannot infer confidently.
Draft multi-entity refactors
Consolidations, migrations, and wide model changes can be packaged into one reviewable ACR instead of dozens of manual edits. A CRM consolidation or platform retirement can come back as one governed package.
Ask live-model questions
Read-only queries skip the change path entirely. Ask "list every integration through our legacy ESB" or "which applications carry PII data objects?" and get the answer from the workspace as it exists right now.
Documentation.
If you want the technical path, these are the four docs that matter first.
Quick start
Create a workspace, authenticate, and get the baseline flow before connecting an agent.
Open guide →MCP setup
Connect Claude Code, Cursor, or another MCP-compatible client to your Albumi workspace.
Open guide →Generation workflow
See how document-driven generation works before you run it against your own material.
Open guide →Governance overview
Review the ACR path, approval flow, and the guardrails around AI-proposed changes.
Open guide →MCP details.
What are the four MCP tools Albumi exposes?
pull_workspace (read the current state - applications, integrations, data objects, capabilities, IT components, organizations, initiatives, and their relationships); validate (run referential integrity and schema validation on a draft package); audit (run the same Architecture Health checks the UI surfaces, returned as structured data); push_workspace (submit a draft package as an Architecture Change Request, never as a direct write).
Can the AI agent change the workspace directly, or only via Architecture Change Requests?
Only via Architecture Change Requests. Push always lands as an ACR, never as a direct write to the live model. The Review Board the ACR is routed to, the voting members, the merge gate - all the same as for human-created ACRs. The agent cannot approve its own ACR.
How is the agent authenticated?
The MCP server runs locally as a CLI; you authenticate once with albumi login against your Albumi account, and the agent inherits your workspace permissions. Whatever you can read or write through the web UI, the agent can read or write through MCP - nothing more.
What MCP transport does Albumi use?
Standard MCP stdio transport. Any MCP-compatible client connects by running the albumi CLI as a subprocess. Claude Code via the plugin, Cursor via .cursor/mcp.json, and any custom MCP client all use the same transport.
What happens if the agent generates broken references or invalid enums?
Validate runs before push, at the workspace level. Orphaned references (entity not found), missing required fields, and invalid enum values are returned to the agent as structured errors - not free text. The agent corrects them and re-validates. Broken state never reaches the live workspace.
Connect an agent to a workspace that can answer back.
Connect your AI agent to a workspace that talks back - included on every plan, no add-on, no premium tier.