Lytho for Claude Connector - MCP Technical Documentation

 

Connect Claude to your Lytho account to run brand compliance checks, manage creative work, route proofs for approval, and search your digital asset library, in conversation, without leaving Claude.

1. Overview

Contents 

Lytho is a content operations platform covering brand compliance, project and proof workflow, and digital asset management. The Lytho connector gives Claude authenticated, permission-scoped access to your own Lytho workspace, so you can ask for work in plain language instead of clicking through the application.

Representative requests once the connector is enabled:

• Check this banner against our Bronze brand pack and tell me what fails.
• Create a proof for the Q3 launch project, capture this file, and put it on the standard marketing review route.
• What proofs are waiting on my review?
• Find every logo asset tagged "primary" in our asset library and give me share links that expire in 30 days.
• Where does the Website Redesign stand, and what is overdue?

The connector operates strictly as the signed-in user. It can see and do exactly what that person can see and do in Lytho, and no more.

1.1 Does the connector read data, write data, or both?

Both. Of the 59 tools available to Claude, 28 are read-only and 31 can change data. Four are annotated as destructive and always prompt for confirmation in Claude. Section 7 marks every tool individually, and Section 9 describes the additional confirmation steps that apply to deletions, bulk updates, and share links.

 

2. Requirements

• A Lytho account on a plan that includes connector access.
• Your Lytho account subdomain, which is the first part of your Lytho web address, typically your company name or an abbreviation of it. You are asked for it during connection.
• Claude on a plan that supports custom connectors, on the web, on desktop, or in Claude Code.
• Digital asset management tools require the Lytho digital asset management add-on. Artificial intelligence brand-compliance tools require the Lytho brand-compliance add-on. If your workspace does not have an add-on, its tools do not appear at all.

Customers who license digital asset management without Lytho Workflow are supported. The connector works and exposes the digital asset management tools.

 

3. Setup

3.1 Server address

https://latus-mcp.wf.lytho.us/mcp

The transport is streamable HTTP, the current Model Context Protocol standard. A standard input and output transport also exists for Lytho's own first-party applications and is not used by Claude.

3.2 Claude on the web and on desktop

1. Open Settings, then Connectors.
2. Choose Add custom connector.
3. Paste the server address above and confirm.
4. Claude opens a Lytho sign-in window. Enter your account subdomain when prompted, then sign in with your normal Lytho credentials. Single sign-on is supported where your organization uses it.
5. Approve the access request. Claude returns to the conversation with the Lytho tools available.

 

3.3 Claude Code

Add the server to your .mcp.json file:

{ 

  "mcpServers": { 

    "lytho": { 

      "type": "streamable-http", 

      "url": "https://latus-mcp.wf.lytho.us/mcp" 

    } 

  } 

} 

The same subdomain prompt and sign-in flow runs on first use.

3.4 Verifying the connection

Ask Claude to list your Lytho projects, or to run the reviewer tool. A successful response confirms authentication, entitlement, and permission resolution in one step. If no Lytho tools appear at all, see Section 12.

3.5 Disconnecting

Remove the connector from Settings, then Connectors in Claude. Access can also be revoked centrally by a Lytho administrator, by disabling the user or the connector client in your identity provider.

 

4. Authentication

The connector uses the OAuth 2.1 authorization code flow with Proof Key for Code Exchange, using the S256 challenge method, against your organization's Lytho identity provider, including single sign-on where configured. Dynamic client registration is supported, so Claude registers itself automatically. Claude never sees or stores your Lytho password.

The connector acts as an authorization proxy in front of the identity provider. It holds no server-side session store and no token database. All transient authorization state is carried inside encrypted, audience-bound, short-lived tokens using AES-256-GCM, and the token issuer is verified on every exchange.

Standard OAuth metadata is published at the following addresses and is discovered automatically by Claude:

https://latus-mcp.wf.lytho.us/.well-known/oauth-protected-resource

https://latus-mcp.wf.lytho.us/.well-known/oauth-authorization-server

No tool in the connector ever asks you for a password, an application programming interface key, or a token. Authentication happens once, through the identity provider's own sign-in page. Treat any request for credentials inside a conversation as illegitimate.

 

5. What Claude can access

Access is decided by your Lytho account, not by the connector. Three layers apply, in order.

1. Product entitlement. Modules your workspace has not purchased contribute no tools. A workspace without the digital asset management add-on sees zero digital asset management tools.
2. Role and permission. Within an enabled module, individual tools are filtered by your Lytho permissions and roles. If you cannot create projects in Lytho, the create_project tool is not offered to Claude, and a call to it is refused.
3. Per-request enforcement. Every tool call is re-authorized on the server against the Lytho application programming interface using your token, including checks that depend on the specific arguments used. Building a review route with tier deadlines, or saving one as a reusable template, each require their own Lytho permission even though the tool itself is visible to you.

 

Visibility filtering is a convenience. The underlying Lytho services remain the authority on every read and every write. Because of this, two people at the same company can see different tool sets in Claude. That is expected and correct.

 

6. Example prompts and expected outcomes

6.1 Check a design against brand guidelines

Prompt. Check if this marketing banner follows our brand guidelines: [image address]

What happens. Claude calls reviewer to discover which compliance packs your account has, for example a Bronze image pack covering typography, tone of voice, and footer, then runs compliance_check_by_pack against that pack.

Expected outcome. A per-check report of which checks passed and which failed, and for each violation the specific problem, its severity, and any suggested fix. A typography failure reports something in the form of Font "Comic Sans" is not approved. Use "Inter" or "Montserrat", rather than a bare failure.

6.2 Create a proof and send it for review

Prompt. I need to submit the Q3 brand refresh assets for review. Create a proof called "Q3 Brand Refresh" and use this file: [address]

What happens. Claude calls get_project_details to find the right project, then create_and_upload_proof to create the proof and capture the file in one step. It calls get_proof_route_templates to list your published approval routes, asks which one you want, and attaches it with attach_proof_route. If you named the route at the outset, it is attached in the original call.

Expected outcome. The proof identifier and a direct link back into Lytho.

6.3 Get the status of a project

Prompt. Give me a status update on the Website Redesign. What is pending, and is anything overdue?

What happens. Claude resolves the project with get_project_details or search_work, then calls project_insights_aims with your actual questions, which runs analysis over the project's real task, assignee, and due-date data.

Expected outcome. A synthesized status update naming overdue items and their dates, with get_lytho_url links to anything worth opening.

6.4 Find and share assets

Prompt. Find every logo asset tagged "primary" and give me a share link that expires in 30 days.

What happens. Claude calls dam_search_assets, then dam_share_asset. Before creating the link, the connector confirms the link name, the expiry, and whether downloads are permitted.

Expected outcome. A copyable share link with the terms you confirmed. No email is sent unless you asked for dam_share_asset_via_email explicitly.

6.5 Review your own queue

Prompt. What proofs are waiting on my review?

What happens. Claude calls get_reviews, which returns your reviewer queue only.

Expected outcome. A list of proofs awaiting your response, with links. Note that recording the review decision itself is not possible through Claude. See Section 9.

 

7. Tool reference

The connector exposes 59 tools to Claude, grouped below. The Access column indicates whether a tool only reads data or can change it. Tools marked Confirms first do not act until you explicitly confirm. See Section 9.

A small number of additional tools exist for Lytho's own first-party applications, namely the browser extension and the Canva and Microsoft Office add-ins. Those are deliberately withheld from third-party artificial intelligence clients and are neither listed nor callable through this connector.

7.1 Brand compliance

Tool	What it does	Access
reviewer	Lists the brand-compliance packs and individual checks configured for your account. Call this first to discover what is available.	Read
compliance_check_by_pack	Runs brand compliance checks on text, an image, or both in one call, either as a whole named pack or as specific named checks.	Read
compliance_request_upload_url	Issues a short-lived, single-purpose upload address so large images or documents can be sent for checking without embedding them in the conversation.	Write, temporary storage only

Available checks. Which checks exist depends on how your account is configured. Ask Claude to run reviewer to see yours. The connector supports typography, logo, colour, footer, tone of voice, text, image quality, content claims, blocklist terms, allowlist terms, promotional language, calls to action, artificial intelligence transparency disclosure, and email regulatory checks. It additionally supports four accessibility checks: alt text, colour contrast, reading order, and readability. Accessibility checks apply to portable document format and presentation files.

Checks are usually grouped into named packs, for example a Bronze pack bundling typography, tone of voice, and footer. Packs differ between image content and text content. Results are returned per check, with each violation carrying a message, a severity, and where applicable a suggested fix.

When a check cannot run, the connector says so plainly. The result is marked as needing manual review, with a retry hint, rather than being reported as a compliance failure. A technical fault is never presented to you as a brand violation.

7.2 Projects, tasks, and campaigns

Tool	What it does	Access
get_project_details	Lists the projects available to you.	Read
get_project_data	Returns the detailed data for a project.	Read
get_project_templates	Lists available project templates.	Read
get_campaigns	Lists campaigns.	Read
get_reports	Lists Workflow reports.	Read
search_work	Searches work items by keyword, including proofs and reviews.	Read
get_work_item	Fetches a single workflow entity by type and identifier.	Read
get_task_detail	Fetches tasks in full detail, including tags, work group, assignees, and per-assignee dates.	Read
get_calendar_events	Reads calendar work items within a date range.	Read
get_lytho_url	Builds a direct link into Lytho for a given item.	Read
create_project	Creates a project.	Write
create_project_with_import	Creates projects in bulk from a comma-separated values import.	Write
create_campaign	Creates a campaign.	Write
create_work_group	Creates a work group container inside a project.	Write
create_task	Creates a task inside a project's work group.	Write
update_workflow_item	Updates fields on a project, task, campaign, or project group. Supports a preview mode that shows what would change without changing it.	Write, destructive, Confirms first
update_calendar_event	Changes start dates, due dates, or both, on a project or task.	Write
delete_workflow_item	Permanently deletes a project, task, or proof.	Write, destructive, irreversible, Confirms first
add_work_item_member	Adds a member to a project or an assignee to a task.	Write
remove_work_item_member	Removes a project member or a task assignee.	Write, destructive
add_campaign_member	Adds a member to a campaign.	Write
remove_campaign_member	Removes a member from a campaign.	Write, destructive
create_comment	Posts a comment on a campaign, project, or task.	Write
update_comment	Edits the text of an existing comment. You can only edit your own.	Write

 

7.3 Proofs and approval routing

Tool	What it does	Access
get_proofs	Lists proofs, optionally filtered by project.	Read
get_reviews	Lists the proofs waiting on your own review, which is your reviewer queue.	Read
get_proof_route_templates	Lists published approval route templates.	Read
create_proof	Creates a proof, which is a review item.	Write
upload_proof	Uploads one or more files to an existing proof.	Write
create_and_upload_proof	Creates a proof and attaches its content in a single step, from uploaded files or by capturing a web address, and can attach an approval route at the same time.	Write
create_proof_route	Builds a full review route on a proof: every reviewer tier in order, optionally with tier deadlines, and optionally saved as a reusable template.	Write
attach_proof_route	Attaches an existing route or route template to a proof.	Write

 

7.4 Project insights

Tool	What it does	Access
project_insights_aims	Runs analysis over a project's real data to answer specific questions you ask about status, risk, overdue work, or bottlenecks.	Read
packaged_insights_aims	Runs a predefined analysis over a project.	Read

 

7.5 Digital asset management

Requires the digital asset management add-on. Individual tools additionally depend on your digital asset management role and on which features your workspace has enabled.

Tool	What it does	Access
dam_search_assets	Searches assets in the asset library.	Read
dam_search_suggest	Returns autocomplete suggestions for asset search.	Read
dam_get_asset_details	Returns full metadata for a single asset: dimensions, size, tags, versions, previews, and custom fields.	Read
dam_download_asset	Returns a time-limited download address and the real file name for an asset.	Read
dam_search_collections	Searches collections.	Read
dam_list_tag_groups	Lists every tag, organized into tag groups.	Read
dam_list_tag_menus	Lists the curated tag menus shown in the Lytho interface.	Read
dam_waiting_room	Lists assets sitting in the waiting room.	Read
dam_approval_queue	Lists assets awaiting your approval, grouped by uploader and date.	Read
dam_upload_asset	Uploads a new asset.	Write
dam_save_to_asset_manager	Promotes waiting-room assets into the main library.	Write
dam_send_for_approval	Submits waiting-room assets for approval.	Write
dam_approve_assets	Approves pending assets, promoting them into the library.	Write
dam_decline_assets	Declines pending assets.	Write
dam_bulk_version_update	Replaces an existing asset's file with a new version.	Write
dam_create_collection	Creates a collection.	Write
dam_save_to_collection	Adds assets to a collection.	Write
dam_share_asset	Creates a copyable share link for one or more assets. No email is sent.	Write, Confirms first
dam_share_asset_via_email	Emails a share link for one or more assets to named recipients.	Write, sends email, Confirms first
dam_update_share_link	Updates an existing share link's name, expiry, or download settings.	Write, Confirms first

How asset search behaves. Searches run as plain keyword matching by default and are ranked by relevance, so an asset that matches only in its description still surfaces first. Ask Claude to use semantic search explicitly if you want it. A search with no search terms is returned newest first, since relevance means nothing without a query. If your workspace has semantic search switched off, a request for it quietly returns keyword results rather than failing.

 

7.6 Utility

Tool	What it does	Access
proxy_image	Fetches an image by web address on the server so it can be used in a compliance check.	Read
get_feature_upgrades	Lists Lytho capabilities available as add-ons that your workspace has not enabled, each with a short description and a link. Returns nothing when you already have everything.	Read

 

8. Resources and prompts

Alongside tools, the connector publishes reference material that Claude can read directly.

Resource	Address	Contents
Workflow Glossary	latus://glossary	Lytho Workflow terminology, so that Claude uses your vocabulary correctly.
Reporting Reference Guide	latus://reporting-reference	A guide to Lytho's reporting concepts.
Server Configuration	latus://config	Server name, version, and enabled modules.

Two reusable prompts are also provided.

Prompt	Purpose
Describe asset	Generates a description for an asset in a formal, casual, or technical style.
Workflow analysis	Analyzes a named workflow for bottlenecks and optimization opportunities.

 

9. Safety behavior

• Every tool declares its risk profile. Read-only tools are annotated as read-only. Tools that write are annotated as not read-only. Tools that destroy data are annotated as destructive. Claude uses these annotations to decide when to ask you before acting.
• Destructive and bulk-update operations require explicit confirmation. delete_workflow_item and update_workflow_item refuse to run until confirmation is given. If Claude attempts one without it, the tool returns a refusal explaining that the action was not confirmed, and the deletion or update does not happen. Nothing is written before the confirmation step, so a refused call leaves no partial change. update_workflow_item additionally supports a preview mode that shows exactly what would change without changing it. Ask for it if you want to see the change first.
• Sharing operations confirm their terms. The share tools establish the link name, the expiry, and the download permissions with you before creating or emailing a link. A link that never expires requires explicit confirmation.
• There is no way to approve or reject a review through Claude. The connector exposes no tool that records a review decision on a proof. This is deliberate. Claude can build routes, attach reviewers, and read your review queue, but approving or rejecting a proof remains something a person does in Lytho.
• Denials are explicit and never upsell. If you lack permission, the connector names the missing Lytho permission, in Lytho's own administrative wording, and tells you to ask your workspace administrator. It never names a product you have not licensed, never suggests a purchase, and never partially completes the work. Discovering what your workspace could add is confined to get_feature_upgrades, which you have to ask for.
• The connector never asks you for secrets. No tool requests a password, an application programming interface key, or a token.

 

10. Known limitations

• Review decisions cannot be recorded through Claude. Approving or rejecting a proof is intentionally out of scope. See Section 9.
• Tool availability varies by user. Two colleagues in the same workspace will often see different tool sets. This follows entitlement, role, and permission, and is not a fault.
• Permission changes require a fresh token. If your Lytho permissions changed recently, disconnect and reconnect the connector so that Claude picks up an updated token.
• Compliance packs are account-specific. There is no universal set of checks. Always run reviewer first to discover what your account has.
• Accessibility checks apply to portable document format and presentation files only.
• Large files are handled by reference, not inline. Use compliance_request_upload_url for large images or documents rather than embedding them in the conversation.
• Uploaded compliance files are temporary. They live in a dedicated upload location and are reachable only through short-lived addresses. They are not retained as library assets.
• The connector has no background access. It acts only during a conversation, and only on your instruction. There is no scheduled synchronization, no crawl, and no pre-indexing.
• Some tools are reserved for Lytho's own applications and are not available through Claude. See Section 7.

 

11. Data handling, privacy, and security

What leaves your Lytho account. Only the data returned by the specific tools Claude calls, in response to your request. The connector does not bulk-export, crawl, or pre-index your workspace, and it has no background access.

Whose services are called. The connector calls Lytho's own first-party services. It does not proxy a third party's application programming interface.

Authentication and authorization. See Section 4 for the authentication design. Every tool call re-validates the caller's token and re-checks entitlement, role, and workspace feature settings before touching a Lytho service, including argument-level checks for capabilities that the Lytho web application itself gates separately. The Lytho services enforce their own permissions independently, so the connector cannot be used to reach data or take actions that the user could not reach in the Lytho web application.

Transport. Encrypted transport is enforced end to end. Cross-origin access is restricted to an explicit list of approved client origins.

File handling. Files sent for compliance checking are stored in a dedicated temporary upload location and reached only through short-lived, presigned addresses lasting minutes. Asset download and share links are likewise time-limited and expiry-controlled.

Conversation data. The connector collects no conversation data beyond the arguments required by the tool being called. It does not read Claude's memory, chat history, conversation summaries, or user files.

Error handling. Upstream failures are reported as failures. Raw exception text from internal services is never surfaced into your conversation. It is logged on the server, and you receive a plain, actionable message instead.

Telemetry. The connector records operational telemetry, namely request timing, tool name, and error rates, for reliability and support. This is used to operate the service.

Retention, sharing, and contact. Lytho's full data processing, retention, and third-party sharing terms, together with the contact details for privacy enquiries, are published in Lytho's privacy policy and terms of service.

Service health. The following endpoints are public and unauthenticated.

https://latus-mcp.wf.lytho.us/health

https://latus-mcp.wf.lytho.us/version

 

12. Troubleshooting

No Lytho tools appear in Claude. Confirm that the connector shows as connected under Settings, then Connectors, and that your Lytho plan includes connector access. If it connected but no tools appear, your account may lack the necessary Lytho permissions. Check with your Lytho administrator.

Some tools are missing, and a colleague has more than I do. Expected. Tool visibility follows your own Lytho permissions and roles, and your workspace's enabled modules. Ask your administrator to grant the relevant permission.

Digital asset management tools are missing entirely. Your workspace does not have the digital asset management add-on, or your user holds no digital asset management role. Ask Claude to run get_feature_upgrades to see what is available to add.

A message says this action requires a permission I do not have. The message names the exact Lytho permission and where an administrator grants it, under Account Settings, then Roles and Permissions. Send that message to your workspace administrator as it is written.

A message says it could not verify my access right now. That is a temporary fault reaching Lytho's permission service, not a permissions problem. Retry in a moment.

The sign-in window asks for a subdomain and I do not know mine. It is the first part of your Lytho web address. If you reach Lytho at https://acme.golytho.us, your subdomain is acme.

Sign-in fails or loops. Sign in to Lytho directly in a browser first, to clear any pending single sign-on prompt, then remove and re-add the connector in Claude.

A tool returns a permission error even though the tool is visible. Some permissions depend on the specific arguments used. Adding deadlines to review tiers, or saving a route as a template, each need their own permission. Permissions are also re-checked at call time against the live Lytho service, which is the final authority. If your permissions changed recently, disconnect and reconnect the connector to obtain a fresh token.

A compliance check came back saying it needs manual review. The check could not complete, usually because of a transient fault in the analysis service. It is not a finding against your asset. Ask Claude to run it again.