Skip to main content
yier_wu
yier_wu
Rookie
July 23, 2026
Question

Expose Knowledge Base tools in the MCP server (read article + list, with link targets preserved)

  • July 23, 2026
  • 2 replies
  • 73 views

Hi Front team,

Sharing a feature request for the MCP server. First, thanks for the open beta and for gathering feedback here directly.


The MCP server currently has no Knowledge Base tools. I'd love to see the KB surfaced through MCP.


We run a public help center on Front's KB and I'm building an AI agent to keep it healthy — specifically a periodic link audit. For each article, the agent needs to look at the outbound links to other KB articles and flag ones that are broken.

Concretely, the tools I'd hope to see
- read_knowledge_base_article — returns body with links as [label](url) (or a structured list of links with hrefs), plus last-modified date
- list_knowledge_base_articles / list_knowledge_base_categories — for enumeration

Later we also like to create drafted articles in some other workflow.
So, would love to know if KB support is on the roadmap, and happy to be an early tester if useful. Thanks!

    2 replies

    martin_silva
    martin_silva
    Fronteer
    July 28, 2026

    Hey ​@yier_wu !

    Thanks for the feedback and for been part of our open beta. Concrete, well-scoped requests like this are exactly what's useful to us to enhance our MCP experience

    The MCP server is deliberately scoped to working conversations as of today — reading, triaging, drafting, and acting on the objects around the inbox (conversations, messages, contacts, accounts, tags, inboxes). It doesn't surface the Knowledge Base at all yet, so there's no way to run the audit you're describing today.

    How we think about it internally: there's a line between an agent operating inside the inbox (what MCP does now) and tools for managing or configuring the workspace itself — the KB, creating tags, setting up rules, and so on. Your link audit is really a KB-management workflow, which sits on that second side. We don't expose that admin/management surface through MCP yet, but it's a direction we're hearing about and actively thinking through and working on it. 

    A few questions to capture this in the right shape if you don’t mind:

    1. Beyond the link audit, what's the bigger goal for the KB agent? keeping the public help center accurate, cutting repeat tickets, feeding cleaner content into AI answers? Curious what "healthy" means to you here. Also, curious on the frequency of this actions and tool calling. This is super important to understand too rate limits
    2. Should the agent just flag broken links (read + report out), or also fix them in Front (write back)? Read-only vs. write changes how we'd scope it.
    3. The "drafted articles in another workflow" you mentioned — what would trigger that, and where are you running the agent today (Claude, a custom setup, something else)?

    We can't commit to timing while KB sits outside the current scope, but this is valuable signal and I'm capturing it directly. And yes — I'd like to take you up on early testing if we open up a KB surface; I'll follow up here if/when that moves.

    Thanks again for helping us shape the beta.

    yier_wu
    yier_wuAuthor
    Rookie
    July 29, 2026

    Hello,

    Thank you for this reply.

    1. Yes, the link audit is just one part of the knowledge agent. We are actively thinking about and testing different use cases and workflows. For example, in a more workflow-support + knowledge scenario, we would like to bridge knowledge gaps identified through support tickets. This would require the MCP to expose tools that allow the agent to modify existing articles, create new ones, etc.
    1. Fixing links is just one of the skills. The bigger picture should involve an agent that can fix different kinds of content that is no longer up to date (broken or outdated links, wording, legacy features, typos, translation errors, etc.).
    2. A support ticket resolution or internal request that identifies a knowledge gap could trigger the creation of a draft article. We are using Claude for this.

    I would be very interested in being involved in the early testing.
    Bests,

    Terms & ConditionsAccessibility statement