Bootstrap your IT map from code with an AI agent (MCP)

Until now, the UrbaHive MCP connector was mostly used to query your information system in plain language. It now does much more: an AI agent can create and update your map. In practice, you can point an agent at your code or architecture documents and let it bootstrap your IT map — applications, flows, servers, processes, capabilities — in minutes, from Claude Code, Cursor, Codex, Windsurf, Devin or any other MCP client.

New to the protocol? Start with What is the Model Context Protocol?. For a guided first connection with Claude Desktop, see Connect Claude to your IT map with MCP.

What changed: read and write

The UrbaHive MCP server now exposes 65 tools: read tools (list_*, get_*, search_architecture, get_architecture_summary) and write tools (create_*, update_*, set_*).

The model stays safe by design:

• Read-only by default. A token is read-only unless you explicitly enable write at creation time (the "write" scope).
• Double barrier. Writing requires both the token's "write" scope and a sufficient role: Owner or Editor for typologies, Owner/Editor/Member for elements. Viewers stay read-only.
• No delete. The protocol exposes no delete tool. Corrections are made in the UI.
• Isolated and audited. A token grants access to a single organization, every call is recorded in your audit log and rate-limited.

For the full security and GDPR picture, see MCP and security: access model, GDPR and IT compliance.

Step 1 — Generate a token with the "write" scope

1. Sign in at app.urbahive.com and open Settings → Connectors.
2. Create a new token, pick the target organization and enable the write option.
3. Copy the token immediately — it is shown only once. Tokens have a lifetime (90 days by default) and can be revoked at any time from the same screen.

The server endpoint is: https://app.urbahive.com/api/mcp. Every client below connects to it with the Authorization: Bearer <PAT> header.

Step 2 — Plug in your agent

Replace <PAT> with your token. Where the format allows it, store the token in an environment variable (e.g. URBAHIVE_PAT) rather than hard-coding it.

Claude Code

`bash

claude mcp add --transport http urbahive \

https://app.urbahive.com/api/mcp \

--header "Authorization: Bearer <PAT>"

`

Then verify with claude mcp list.

Claude Desktop / claude.ai

Ready-to-paste config lives right inside Settings → Connectors in UrbaHive: Claude Desktop uses the mcp-remote bridge, and claude.ai accepts the URL via *Add custom connector* (URL + Bearer PAT).

Cursor — ~/.cursor/mcp.json

`json

{

"mcpServers": {

"urbahive": {

"url": "https://app.urbahive.com/api/mcp",

"headers": { "Authorization": "Bearer ${env:URBAHIVE_PAT}" }

}

}

}

`

Windsurf — ~/.codeium/windsurf/mcp_config.json

`json

{

"mcpServers": {

"urbahive": {

"serverUrl": "https://app.urbahive.com/api/mcp",

"headers": { "Authorization": "Bearer ${env:URBAHIVE_PAT}" }

}

}

}

`

Codex CLI — ~/.codex/config.toml

`toml

[mcp_servers.urbahive]

url = "https://app.urbahive.com/api/mcp"

bearertokenenvvar = "URBAHIVEPAT"

`

Devin

In Devin's MCP settings, add a Streamable HTTP server with URL https://app.urbahive.com/api/mcp and header Authorization: Bearer <PAT>.

Any other client (stdio-only)

Clients that only speak stdio can bridge through mcp-remote:

`bash

npx -y mcp-remote https://app.urbahive.com/api/mcp \

--header "Authorization: Bearer <PAT>"

`

Step 3 — Bootstrap your map from code

With a "write" token, hand the agent your repository or architecture docs and let it create items in the right dependency order:

1. Typologies — create_editor, create_infrastructure, create_flow_type, create_project_advancement_status, create_dat_version_status, create_organization_role.
2. Programs — create_program.
3. Applications, servers, flows — create_application, create_server, create_flow.
4. Data objects — create_dat_object, create_dat_version.
5. Processes — create_process_domain, create_process, create_process_step, then set_process_step_actors, set_process_step_applications, set_process_relations.
6. Capabilities — create_capability, then set_capability_applications.

Example prompt: *"Read this repo, infer the applications, data flows and servers, then create them in UrbaHive via the MCP tools, following the typology → program → application → flow order."*

A few guardrails:

• Creating typologies requires an Owner or Editor role; a Member can create elements but not typologies.
• Since no delete tool is exposed, fix any mistakes in the UI.
• Ask the agent to work in batches and verify after each step (using the list_* tools) to avoid duplicates.

Once the base is seeded, you get all of UrbaHive: coverage matrix, impact analysis, risk detection. To go further, see Enterprise mapping: a complete guide for SMBs and Link business processes to your application map.

FAQ

Do I need to code to bootstrap my map from code?

No. The AI agent reads the code and calls the tools. Your job is to provide context (repo, diagrams, inventory) and review the result in UrbaHive.

Which clients are supported?

Any "streamable HTTP" MCP client: Claude (Desktop, Code, claude.ai), Cursor, Windsurf, Codex, Devin, and others via mcp-remote.

Can an agent delete data by mistake?

No. No delete tool is exposed, and writing requires a "write" token plus an adequate role. When in doubt, keep a read-only token.

Does data leave the EU?

No. The MCP server is hosted in Europe and every call is logged. See MCP and security.

Start for free at app.urbahive.com/signup and explore the connector on urbahive.com/connectors.