Your brand is now a tool any AI agent can use, we just shipped Markolé MCP

When we built Markolé, the pitch was simple: build your brand identity through AI-guided interviews, and from that point forward every document, image, and asset is generated on-brand by design, not by luck.

A few weeks ago we open-sourced our JSON-LD brand schema so you could take a snapshot of your brand into any AI tool. The reaction was great, but the next question was obvious:

"What if my agent could just talk to Markolé directly?"

This week, that's exactly what shipped.

Markolé is now an MCP server

Model Context Protocol (MCP) is the open standard for letting AI agents read from and write to external systems through a typed, structured interface. It's what makes Claude Desktop able to read your Gmail, edit your Notion, or query your database.

Markolé now speaks MCP natively. That means:

Claude Desktop, Claude Code, and any other MCP-aware agent can connect directly to your Markolé brand with a personal access token.
The agent doesn't see a flattened PDF. It sees the same structured brand DNA Markolé uses internally — narratives, traits, visual identity, presentations — all of it, with proper semantic labels.
MCP is a paid feature. Guided users get read-only access, so any external agent can pull the structured brand and stay on-brand. Pro users get the full write surface — designed specifically for agencies, brand consultants, and brand teams who want to treat Markolé as their structured-data brand repository, and get the instant benefit of Markolé's tooling on top: logo generation, brand imagery, presentations, narrative synthesis.

This is bigger than the JSON-LD export. The export is a snapshot. MCP is a live, two-way connection.

What agents can actually do

The read surface gives an agent the full picture of any brand the user has access to:

List brands and teams
Read brand summaries, narratives, traits, and visual identity
Pull the full brand book or final report as JSON-LD
Read the visual identity as JSON-LD

The Pro write surface is lock-aware and built for agency-grade workflows:

update_brand_narrative — rewrite a vision, mission, value, archetype, or any of the 17 narratives
update_visual_identity — adjust color palettes, typography, dos & don'ts
upload_logo_asset — drop in an existing logo from a data URL or public image URL
generate_logo_asset — have Markolé generate logomark, logotype, and dark variants directly, with the creative brief stored for future regenerations
generate_brand_imagery — produce primary 3:2 brand imagery using full brand context (existing logos, imagery, product references all feed in)
update_brand_presentation — drop a complete presentation; slide background images regenerate automatically when the brief changes, and the old ones are deprecated cleanly
update_final_brand_report and update_final_visual_report — write back synthesized brand content
list_writable_teams and create_brand — autonomous agents can create entirely new brands in any team they're an admin or editor of, defaulting to the user's personal team

Every write respects the section locks users set inside Markolé. If you lock your visual identity, an agent that tries to modify it gets a structured 423-style error listing the exact locked sections and clear unlock guidance — not a silent overwrite.

And because Markolé's realtime sync stays on, any update from an MCP client immediately shows up for connected teammates with the right lock indicators and refresh states.

Why this matters for agencies and brand consultants: instead of juggling PDFs, decks, and ad-hoc prompts across clients, you point your agent at Markolé as the source of truth. The agent reads structured brand content, writes back updates, generates production-ready logos and imagery — and your client sees every change live in the same workspace.

Deep web brand discovery — agentic crawling, baked in

Alongside MCP, we shipped something we've wanted for a long time: a "Discover from the web" workflow inside the Brand Analysis tab.

You give us your brand name, your industry, and optionally up to three competitors. Markolé then:

Searches the web for each entity using Gemini's google_search grounding.
Crawls up to 75 URLs across a depth of 3, using url_context to read each page and pick the most relevant child links.
Extracts brand-identity content — mission, values, tone, positioning, archetype signals — and runs every chunk through the same trait and topic extraction prompts we use in the guided interview.
Streams results live: a 3-line sanitized preview as snippets land, a Sub-Category bubble chart that updates in realtime, and a running count of URLs analyzed.

Every trait pulled from the web lands in your brand's evidence log with full provenance — source URL, source type, depth, which entity it came from (your brand or a competitor), and which run produced it. Competitor evidence merges into the same DNA pool so you can see contrast and overlap, but the audit trail is always there.

The reason this matters: most brand identity content isn't on the homepage. It lives on /about, /values, /sustainability, press pages. A depth-3 crawl is what surfaces that — and Markolé does it without you ever leaving the Brand Analysis tab.

A big speed bump across the platform

With the recent Gemini Flash 3.5 release, we took the opportunity to rebalance our model selection across Markolé. High-volume work — synthesis, summaries, topic extraction, the brand chat — now routes through the model tier that fits each task best.

The result: a huge speed increase across the app. Synthesis screens feel snappier, chat replies come back faster, and per-brand cost dropped at the same time. No quality regression where we measure it.

Deprecated images, now searchable and saveable

When you regenerate a logo, brand image, or marketing asset, Markolé deprecates the old version. Until now, those old versions just quietly aged out.

This week we added a Deprecated Images gallery at the bottom of the Marketing Documents image area. It surfaces every deprecated image for the current brand in a large masonry layout, with:

Open in viewer
Download (so you can keep a version we were about to clean up)
Delete individually
A "Delete all" button if you want a hard cleanup

Deprecated images are automatically deleted after 30 days. The gallery is your chance to rescue anything you still want before that happens.

Settings, cleaned up

Settings had become a junk drawer: identity, brands, teams, company profiles, guidance preferences, MCP tokens, billing — all in one page.

This week we split it:

Profile holds everything about you: account, security, brands, teams, company profiles.
Settings stays focused on behavior and integrations: guidance preferences, MCP access, and billing.

Plus a new MCP Access card with token creation and revocation, a token table, tier-aware messaging in EN and FR, and a copy-ready Claude Desktop config snippet — collapsible, with Copy always visible.

What's next

We think structured, machine-readable brand identity is going to become standard infrastructure as more of the content layer moves to AI agents. The JSON-LD export was step one — describing your brand to an agent. MCP is step two: letting the agent work with it under the same rules and guarantees Markolé itself enforces.

If you're on Pro and you want to wire your brand into Claude Desktop or Claude Code, the Settings → MCP Access page has everything you need. Guided users get the same page with read-only tokens.

Give it a spin on Markolé and tell us what your agents build.