DeployIt blog — product knowledge for humans and AI agents
Guides on MCP servers, wiring support agents like Intercom Fin or HubSpot Breeze to your codebase, and keeping product knowledge — docs, changelogs, help centers — up to date at every deploy.
New here? Start with the MCP server for your codebase, or see how the same expert answers your team in the AI Help Center & Messenger.
How do you know which of your AI agent's answers changed after a release?
Replay a fixed list of questions after every release and compare with the previous run. The most dangerous answer is the one that should have changed and did not move.
Read more →How do you test an MCP server before agents start using it?
Test an MCP server in two passes: the protocol with the Inspector, then the answers with a fixed sheet of questions that includes the ones it must refuse.
Read more →How do you change an MCP server without breaking the agents already using it?
An MCP tool has no version field. Change it safely by adding first, warning in the description, and removing last. A rename traced over four dates.
Read more →What should an AI agent refuse to answer about your product?
An agent should refuse whatever the code cannot prove: roadmap, unpublished prices, commitments. A sorting test of eight questions shows where the line falls.
Read more →Why does your AI agent say a feature doesn't exist when it shipped last week?
A denial of a feature that exists has one of four causes. Four questions, asked in order, tell you which one you are looking at, and what to change.
Read more →How do you write MCP tool descriptions an agent actually uses correctly?
An agent picks an MCP tool from its description and nothing else. Four rewrite passes turn a vague entry into one that gets called for the right questions and skipped for the wrong ones.
Read more →How do you publish release notes in several languages without retranslating them every release?
Retranslating release notes each release makes the languages drift apart. The way out: one source of facts, translations that carry only the form, and the same set of facts in every language.
Read more →Why do your support macros go stale after every release?
A support macro is an answer written by hand on a given day: the next release makes it wrong and nobody reopens it. A post-mortem of an illustrative macro, and the routine that keeps macros current.
Read more →What happens in the first week an AI agent is connected to your product?
The first seven days of an agent connected to your product: the questions that land, the ones that miss, the ones left unanswered, and what you adjust on your side. An illustrative scenario with no usage figures.
Read more →Why does an AI agent answer differently on staging than on production?
Same agent, same question, two answers: staging and production are not the same product, and the agent answers from whichever state its source reflects. Four sources, four ages, and the question to ask before you connect one.
Read more →What security questions should you ask an MCP server vendor?
Twelve questions, in four blocks, that a security team can put to any MCP server vendor: who can call it, what it can read, what it can do, and what you can prove afterwards. For each, the answer that closes the point and the one that leaves it open.
Read more →How do you connect Cursor or Windsurf to your product knowledge?
Both editors accept a remote MCP server from one JSON entry. Here are the two entries side by side, as the docs stated them on 13 October 2026, and what an editor can ask a product expert that it cannot ask the open file.
Read more →What should you expose when you build an MCP server on your own SaaS?
The catalog of an MCP server is a product decision, not an export of your API. Three questions sort what an outside agent should be able to call.
Read more →Why does my AI agent give wrong answers about our API after a release?
An agent plugged into your API documentation answers with the API as it was last documented. After a release, the gap is invisible to it. What to hand it instead.
Read more →Where does the data go when an AI agent asks questions about your product?
A question to a product-knowledge server travels through four places: the Git provider, the vendor's infrastructure, a language-model provider, and the platform that asked. Self-hosting your portal settles the first two at most. Here is what DeployIt's public DPA says about each, and what it leaves open.
Read more →How does an AI agent authenticate to an MCP server?
An agent authenticates to a remote MCP server with an OAuth access token, obtained through a flow the server starts by answering 401. What that token can do, and what revoking it actually cuts, depends on which of three credentials you mean.
Read more →How do you ask your codebase questions via an API?
An MCP server in front of your codebase turns a plain-language question into a sourced answer. The real shift: the caller is no longer a person, and nobody proof-reads before the answer gets used.
Read more →How do you give AI agents access to your code safely?
The safe model is not opening your repository to an agent. It is opening a place that answers questions about it. One access grant followed across five moments, and the four properties tested at each.
Read more →Why does our AI agent hallucinate product features?
An agent invents a feature because it has to answer and nothing forces it to check first. The four shapes that invention takes, and the condition missing when a connected source fails to stop it.
Read more →Claude project knowledge vs an MCP server: which one knows what changed?
Project knowledge is a folder of files, frozen at your last click on Sync, with no history. An MCP server is a source queried at the moment the question is asked. One week in a repository shows the difference.
Read more →GitHub MCP server: how to set it up, and what it can actually answer
Setup in three minutes, the list of what it answers well, and the call log that shows where it breaks down: reading a diff is one call per commit, and it is not the default.
Read more →Can you connect HubSpot Breeze to your codebase?
Not the way you connect Intercom Fin. Breeze only talks to MCP servers HubSpot has vetted — here is the dated record of what the docs allow, and the two routes that actually work for grounding a Breeze agent in your code.
Read more →10 minutes now. The dashboard tonight, your first report on Monday. You decide.
Free plan, no credit card. Read-only. No action required from your team.
Free plan · No CC · Read-only · Hosted in EU