---
title: "Best MCP servers for user research 2026: 10 compared"
description: "We compared 10 MCP servers that let an AI assistant run interviews and surveys or read the feedback you already have: tools, sign-in, free plans, prices."
url: "https://raconte.ai/en/blog/top-mcp-servers-user-research"
---

# The best MCP servers for user research and customer feedback in 2026

We compared 10 MCP servers that let an AI assistant run interviews and surveys or read the feedback you already have: tools, sign-in, free plans, prices.

November 4, 2026

Godefroy

Key takeaways

*   MCP servers for user research do one of two jobs: start new research from the assistant, such as an interview or a survey, or read feedback that already exists in a research repository, a support inbox or a feedback analytics tool.
*   Among the ten servers ranked here, four let an assistant run a real conversation with a user and read the transcript back: Raconte, Perspective AI, User Intuition and UserCall.
*   Most of these servers ask you to sign in with OAuth. Several also accept an API key, which is simpler for scripts, n8n and headless agents.
*   The MCP connection itself is almost always free. You pay for the product behind it: Tally and Dovetail have a free plan, User Intuition charges $30 or $60 per interview, and Enterpret sells on quote.
*   Read-only servers such as Userbrain and Enterpret are useful for synthesis, but they cannot ask anyone a new question.

The best MCP server for user research depends on whether your assistant needs to ask new questions or read the answers you already have. To run an interview from Claude, Cursor or another AI assistant and get the transcript back, use Raconte, Perspective AI, User Intuition or UserCall. To send a form or a survey, Tally and Survicate both publish an official server. To search what users already told you, Dovetail reads a research repository, Intercom reads support conversations, Userbrain reads usability tests and Enterpret queries feedback gathered from dozens of channels. This page compares the ten servers on what the assistant can do with them, how many tools they expose, how you sign in, and what it costs to start, with every detail read in each vendor’s documentation in November 2026.

Raconte publishes this article and is one of the ten products it ranks. We put our own server first, so we publish the rule behind the order for you to check and disagree with. Every entry links to the vendor’s site and to its MCP documentation where it has one.

Generic developer servers, such as GitHub, Playwright, Context7 or a filesystem server, fill most “best MCP servers” lists. They help an assistant write and test code. None of them talks to a user, so they are left out here.

## How we compared them

An MCP server, short for Model Context Protocol server, is a set of tools a product exposes so an AI assistant can use it directly: create something, read something, change something. We read the MCP documentation, the pricing page and the help center of each vendor in November 2026, and kept only official servers published by the vendor itself. Community servers built by third parties are left out, because nobody guarantees they will keep working.

We compared the servers on seven criteria:

*   what the assistant can do: start new research, read existing feedback, or both
*   whether the assistant gets back the participant’s own words, a full transcript or the raw answers, or only a summary
*   the number of tools, when the vendor publishes it
*   how the client signs in: OAuth in the browser, an API key, or either
*   where the server runs: a remote endpoint, a package you install locally, or both
*   the clients the vendor names, such as Claude, ChatGPT, Cursor or VS Code
*   the price to start: a free plan, a free allowance at signup, a trial, or a quote

The ten servers fall into two families that answer two different questions, so each family has its own ranking. Inside the first family, servers that hold a conversation come before those that send a form, because a conversation records the reasons behind an answer, which is what most research is after. A team that wants scores from a large sample should read that family the other way round. Each of the three groups (conversation, forms, reading) then follows the same order: a free allowance that renews every month, then a one-off free allowance, then a trial, then a quote. Ties go to the larger one-off allowance and to the trial with the lower published price. That order suits a small team that wants to run research from its assistant this month. A team that ranks on panel size, languages or analysis features would put User Intuition or Dovetail higher.

Server

What the assistant does

Tools

Sign-in

Where it runs

Free to start

Paid from

[Raconte](https://raconte.ai/en)

Runs voice interviews, reads full transcripts

8

API key

Remote

60 interview minutes a month

$0.12 per minute of speech

[Perspective AI](https://getperspective.ai)

Runs voice or text conversations, reads transcripts

40 listed

OAuth or workspace key

Remote

100 credits at signup

$99 a month

[User Intuition](https://www.userintuition.ai)

Plans studies, recruits, runs interviews, reads transcripts

Not stated in its guide

OAuth or API key

Remote or local

3 voice interviews at signup

$30 per voice interview

[UserCall](https://www.usercall.co)

Creates voice interview studies, reads themes and transcripts

17

OAuth, or API key locally

Remote or local

First interview

$89 a month

[Tally](https://tally.so)

Creates and edits forms, reads submissions

Not published

OAuth or API key

Remote

Free plan, unlimited forms

€20 a month

[Survicate](https://survicate.com)

Drafts surveys, reads results

18

OAuth only

Remote

10-day trial, 25 responses

$114 a month, billed yearly

[Dovetail](https://dovetail.com)

Searches and edits a research repository

50

OAuth or API token

Remote or self-hosted

Free plan

Quote

[Intercom](https://www.intercom.com)

Reads support conversations and contacts

14, plus 5 for Fin

OAuth or API token

Remote, US or EU

14-day trial

$29 per seat a month

[Userbrain](https://www.userbrain.com)

Reads usability tests and their insights

5, read-only

OAuth

Remote

30-day trial, 2 testers

$49 a month

[Enterpret](https://www.enterpret.com)

Queries feedback from 50+ channels

7

OAuth or bearer token

Remote

None

Quote

## MCP servers that start new research

These six servers let an assistant ask users something new. Four of them run a conversation in which an AI asks follow-up questions. The other two create forms or surveys that a person publishes and shares.

### 1\. Raconte

![Screenshot of the Raconte homepage](/.netlify/images?url=_astro%2Fhomepage-en.evuUlE9t.png&w=1920&h=1200&dpl=6b031bf5959953076155d961)

[Raconte](/en) runs voice interviews with an AI. A person opens a link in the browser, with no account and nothing to install, and talks for a few minutes. The AI asks a follow-up question when an answer stays vague. Once the call ends, the team gets the transcript, the audio, a summary and a sentiment on each answer.

Its [MCP server](/en/docs/mcp) exposes eight tools over Streamable HTTP at `https://api.raconte.ai/api/mcp`, four for interviews and four for invitations. `create-interview` writes the interview from a prompt and, given a list of invitees, emails each person a link. `get-interview` returns every invitation of an interview with its full transcript, its summary and a sentiment on each message, in one call. The other tools list, edit, cancel, reopen and archive interviews and invitations. The handshake and the tool list need no key, so a client can read what the server does before it has one. Raconte gives [setup steps for Claude Desktop, Claude Code, Cursor, Opencode, VS Code with GitHub Copilot Chat and Windsurf](/en/guides/mcp-setup).

Raconte uses an organization API key rather than an OAuth flow, so there is no Connect button: a person creates the key in the settings and pastes it into the client. The same key works in a header for scripts, n8n or an agent running unattended, and in the URL for Claude Desktop, whose custom connectors take a URL and nothing else. It also works with the [REST API](/en/docs/api), the webhooks, the CLI and the Node SDK, all [documented for developers](/en/developers) with a public OpenAPI spec.

The first 60 interview minutes each month are free, with no card. After that, Raconte charges $0.12 per minute of active speech, silences excluded, with top-ups from $12 and no subscription, so a ten-minute interview costs about $1.20. Interviews run in English and French only. In exchange, Raconte picked each voice in those two languages one by one and lists them all in a public catalog, where a team can listen to them before sending anything. Raconte has no participant panel, so a team interviews the customers and users it already has, with no recruiting fee on top of the minutes. The application and the database are hosted in the European Union. A team can also run every AI step with European providers, transcription and voices included, by turning off one organization setting. Raconte produces no report across several interviews. An assistant gets every transcript of an interview from `get-interview` instead, word for word with its sentiment, and compares them itself.

### 2\. Perspective AI

![Screenshot of the Perspective AI homepage](/.netlify/images?url=_astro%2Fperspective.DCs_dhkn.png&w=1920&h=1080&dpl=6b031bf5959953076155d961)

[Perspective AI](https://getperspective.ai) was founded in 2024 in the Bay Area by Guy Nirpaz, who previously founded the customer success platform Totango, and raised a $4M seed round in January 2025. Its agents hold a conversation by voice or text, ask follow-up questions in real time and pull structured data out of the answers. It sells them as a replacement for static forms, for intake and qualification as well as research.

Its [MCP server](https://getperspective.ai/docs-markdown/build/mcp) is the broadest of the conversation tools here. The documentation lists 40 tools in 15 families, labelled as examples. An assistant can create a study, invite participants by email, run outreach campaigns, read conversations word for word, search across them and run analysis on the results. Sign-in is either OAuth 2.1 with dynamic client registration, which most clients handle on their own, or a workspace-scoped key for scripts and CI jobs. Perspective AI lists one-click setup in Claude, ChatGPT, Cursor and Grok, and command setup in Claude Code, Codex, VS Code and Gemini CLI.

MCP access comes with every plan. The free plan gives 100 credits once, at signup. One conversation uses about 10, so the free plan covers about ten conversations. Pro costs $99 a month for 1,000 credits, about a hundred conversations. Perspective AI claims more than 100 languages.

### 3\. User Intuition

![Screenshot of the User Intuition homepage](/.netlify/images?url=_astro%2Fuserintuition.OO7oaucf.png&w=1920&h=1080&dpl=6b031bf5959953076155d961)

[User Intuition](https://www.userintuition.ai) was founded in 2025 in Austin, Texas. It runs research end to end: it plans a study, recruits participants from its own panel or invites yours, runs AI-moderated interviews by voice, chat or video in more than 80 languages, and writes the report.

Its [MCP server](https://www.userintuition.ai/reference-guides/user-intuition-mcp-server/) covers the same steps. An assistant can draft the discussion guide and screener, estimate and launch a panel study, pause fielding, open transcripts and recordings, read reports whose quotes link back to each interview, and generate decks or highlight videos. The guide gives no tool count but offers reduced tool profiles for agents that need fewer tools. The hosted endpoint at `https://mcp.userintuition.ai/mcp` accepts OAuth for interactive clients and an organization API key for headless ones. A local npm package is also available. The company names Claude, ChatGPT, Claude Code, Codex and Cursor.

Signup includes three free voice interviews with your own participants. After that, a voice interview costs $30 with your own participants or $60 with panel recruitment, with no monthly fee and no extra charge for the MCP connection. Professional costs $2,499 a month with a three-month minimum. User Intuition is the only conversation tool here that recruits participants from the assistant.

### 4\. UserCall

![Screenshot of the UserCall homepage](/.netlify/images?url=_astro%2Fusercall.CJ7nUgNA.png&w=1920&h=1080&dpl=6b031bf5959953076155d961)

[UserCall](https://www.usercall.co) was founded in 2024 by Junu Yang, previously a designer at IDEO. Its AI moderator runs voice interviews and decides in the moment whether to probe or move on, then groups the answers into themes, each linked to the quotes it came from.

Its [MCP server](https://app.usercall.co/docs/mcp) has 17 tools. Eight manage studies: list, create, update, review the guide, simulate an interview as a dry run, check status, read results and delete. The other nine manage research triggers, which invite a user when a product event fires in PostHog, Mixpanel, Amplitude or Segment. The server returns themes, insights and risks by default, and the full transcripts and audio links on request. UserCall leaves two steps to a person on purpose: sharing the interview link, and switching a trigger on from the activation page. No tool does either. The hosted connector at `mcp.usercall.co` uses OAuth, while a local package and a REST Agent API take an API key, at 60 requests a minute. Claude, ChatGPT, Cursor and Codex are named.

The first interview is free. Lite costs $89 a month for 100 credits, which UserCall says covers ten to twenty interviews. Its event triggers suit a product team that already tracks cancellations or signups in an analytics tool.

### 5\. Tally

![Screenshot of the Tally homepage](/.netlify/images?url=_astro%2Ftally.FGwBFfdF.png&w=1920&h=1080&dpl=6b031bf5959953076155d961)

[Tally](https://tally.so) was founded in 2020 in Ghent, Belgium, by Marie Martens and Filip Minev. Its builder works like a text document: you type the questions and the form takes shape as you write.

Its [MCP server](https://tally.so/help/mcp) at `https://api.tally.so/mcp` creates forms from a description, edits fields and settings, searches the workspace and fetches submissions for analysis. It cannot delete a form or a submission. Tally recommends OAuth, but an API key works too. Tally names Claude, ChatGPT, Grok, Cursor, Windsurf, Zed, Cline and Continue, and does not publish a tool count.

The server is free on every plan, including the free one, which already allows unlimited forms and submissions within fair use. Pro costs €20 a month. Follow-up questions come from the conditional logic you set up, in a form the respondent fills in by typing. Tally is the obvious place to start for a form an assistant can build and read without any budget.

### 6\. Survicate

![Screenshot of the Survicate homepage](/.netlify/images?url=_astro%2Fsurvicate.DxdkGrwK.png&w=1920&h=1080&dpl=6b031bf5959953076155d961)

[Survicate](https://survicate.com) was founded in 2013 in Poland. It sends short surveys by email, by link, on a website and inside web and mobile apps, and adds AI follow-up questions to typed answers.

Its [MCP server](https://developers.survicate.com/mcp/technical-overview/) at `https://mcp.survicate.com/` has 18 tools. An assistant can list surveys, read results question by question, ask a question across all feedback, draft a survey and archive surveys. Eight more tools write to its Research Hub, such as creating a research project or a note. A drafted survey waits for a person to publish it. Sign-in is OAuth 2.1 only, with no API keys. Survicate names Claude, ChatGPT and Copilot.

There is no free plan. A 10-day trial includes 25 responses. The pricing page lists MCP on every paid plan, starting with Growth at $114 a month billed yearly. Survicate suits a team that wants many short answers and lets the assistant read the open ones.

## MCP servers that read the feedback you already have

These four servers give an assistant access to research and feedback already collected: interview notes, support tickets, usability test videos, survey answers. None of them asks a user anything new.

### 7\. Dovetail

![Screenshot of the Dovetail homepage](/.netlify/images?url=_astro%2Fdovetail.CTgWHaBH.png&w=1920&h=1080&dpl=6b031bf5959953076155d961)

[Dovetail](https://dovetail.com) was founded in 2017 in Sydney and is the best-known research repository. Teams store interview recordings, transcripts, highlights and insights there, and tag them across projects.

Its [MCP server](https://developers.dovetail.com/docs/mcp) is the most complete in this family, with 50 tools. An assistant can search the workspace, read documents and data, create highlights on a transcript, write docs, apply tags and import files. The remote endpoint at `https://dovetail.com/api/mcp` accepts OAuth or a Dovetail API token. A self-hosted server covers clients that only run local processes. Dovetail names ChatGPT, Claude, Figma, Copilot, Cursor and Windsurf.

Dovetail has a free plan with one project and one channel. The paid plan above it is on quote. The MCP documentation does not say which plans include the server. Atlassian, Notion and Canva are among its named customers.

### 8\. Intercom

![Screenshot of the Intercom homepage](/.netlify/images?url=_astro%2Fintercom.DS8E2goC.png&w=1920&h=1080&dpl=6b031bf5959953076155d961)

[Intercom](https://www.intercom.com) was founded in 2011 and sells a customer support platform, with Fin, its AI agent, answering customers before a person does. A support inbox holds what customers complain about without being asked, which makes it a source of feedback many teams never search.

Its [MCP server](https://developers.intercom.com/docs/guides/mcp) has 14 tools: search and read conversations, contacts and companies, read and write help articles, and add an internal note. It cannot send a reply a customer would see. Workspaces with Fin Operator get five more tools over OAuth. Endpoints exist for US and EU workspaces, but not yet for workspaces hosted in Australia. Sign-in is OAuth or an API token. Claude, Cursor, Windsurf and VS Code are named.

Intercom offers a 14-day trial. Essential costs $29 per seat a month, plus $0.99 for each conversation Fin resolves.

### 9\. Userbrain

![Screenshot of the Userbrain homepage](/.netlify/images?url=_astro%2Fuserbrain.D1WLnAvB.png&w=1920&h=1080&dpl=6b031bf5959953076155d961)

[Userbrain](https://www.userbrain.com) is an Austrian company, based in Graz, that runs unmoderated usability tests: testers from its panel record themselves while they try a website or a prototype, and the team watches the videos.

Its [MCP server](https://help.userbrain.com/help/userbrain-mcp) at `https://mcp.userbrain.com/mcp` exposes five read-only capabilities: list tests, look up a test and its tasks, read insights, review tasks and list tester sessions. The documentation states that it cannot create a test or order testers. Sign-in is OAuth with your own Userbrain account. ChatGPT, Claude, Claude Code and Microsoft Copilot are named.

A 30-day trial includes two panel testers, with no card. Starter costs $49 a month for three testers and one test set-up.

### 10\. Enterpret

![Screenshot of the Enterpret homepage](/.netlify/images?url=_astro%2Fenterpret.Pr0gapqG.png&w=1920&h=1080&dpl=6b031bf5959953076155d961)

[Enterpret](https://www.enterpret.com) was founded around 2020 in San Francisco. It gathers feedback from more than 50 channels, such as support tickets, app store reviews, surveys and sales calls, and organizes it into a knowledge graph of themes.

Its Wisdom [MCP server](https://helpcenter.enterpret.com/en/articles/12665166-wisdom-mcp-server), live since December 2025, has seven tools: read the organization details and the graph schema, search fields and values, run a query on the graph, and find a quote from a user. An assistant can ask how many customers mentioned a problem last month and pull the quotes behind the number. The endpoint at `https://wisdom-api.enterpret.com/server/mcp` accepts OAuth or a bearer token. Each person authenticates with their own account. Claude, Claude Code, Cursor, ChatGPT, Codex, Notion and Glean are named.

Pricing is on quote, with no free plan. Canva, Notion and Perplexity are among its named customers.

## Other servers we looked at

Fourteen more products publish an official MCP server close to this topic. They are left out of the ranking for lack of room, a sales-only offer or a narrower fit.

*   Sprig, SurveyMonkey, Typeform and Jotform let an assistant create surveys or forms and read the responses. Sprig sells only after a demo, SurveyMonkey’s server works only for accounts with data stored in the US, and Typeform and Jotform accept OAuth only.
*   PostHog includes survey tools in its product analytics server.
*   UserTesting launched an early-access server in September 2026. It creates and launches tests with its panel.
*   Feedbk deploys AI-moderated interviews and surveys from an assistant, with no public price.
*   Great Question recruits, schedules interviews and searches its repository, on its Enterprise plan only.
*   Maze and Condens read research studies and transcripts. Maze’s server comes with its Enterprise plan, and Condens needs its Business plan or above.
*   Chattermill, Zonka Feedback and BuildBetter read feedback analytics or call transcripts.
*   Pendo reads and writes in-app feedback and survey scores.

Productboard and Zendesk had only community servers when we checked, so they are not on this list.

## How to choose between them

Start from what the assistant is missing.

*   It needs to hear from users it has never asked. Take a conversation server. Raconte suits a team that already has the people to interview, wants to pay by the minute and needs English or French. It gives the assistant every transcript to compare, rather than a report across interviews. Perspective AI has the most tools and runs conversations by voice or text. User Intuition recruits from a panel in more than 80 languages, at $60 per interview with recruitment ($30 with your own participants). UserCall starts interviews from product events.
*   It needs a score or a quick answer from many people. Take a form or a survey server. Tally costs nothing to start. Survicate has no free plan but also reaches people inside the product.
*   It needs to search what users already said. Take a reading server: Dovetail if your research lives in a repository, Intercom for support conversations, Userbrain for usability test videos, Enterpret to query every channel at once.

Many teams connect one server of each kind. An assistant reads support tickets in Intercom or themes in Enterpret, spots a problem nobody understands yet, then creates an interview with Raconte and invites the customers who raised it. A day later, it reads the transcripts and writes up what they said. [Agents that work from what users actually said](/en/blog/context-engineering-ai-agents) make fewer guesses. This loop needs no meeting, yet it covers the step teams most often skip in [human in the loop automation](/en/blog/human-in-the-loop-automation): asking the users themselves.

Check how the client signs in before you choose. OAuth gives each person their own sign-in and works with one-click connectors in Claude and ChatGPT. An API key suits an agent running unattended, a script or an n8n workflow, where nobody is there to click. Raconte uses a key only. Perspective AI, User Intuition, Tally, Dovetail, Intercom and Enterpret accept both. UserCall uses OAuth for its hosted connector and a key locally. Survicate and Userbrain use OAuth only.

If you would rather compare the products than their MCP servers, we also ranked [user interview platforms](/en/blog/top-user-interview-platforms) and [customer feedback tools](/en/blog/top-customer-feedback-tools).

## Frequently asked questions

### What is an MCP server?

An MCP server is a set of tools that a product exposes through the Model Context Protocol, an open standard Anthropic published in November 2024, so that an AI assistant can use the product directly. Once a client such as Claude, Cursor or ChatGPT is connected, the assistant reads the list of tools and calls the right one when you ask in plain language, for example to create an interview or to fetch survey answers.

### What are the best MCP servers for user research?

To start new research from an assistant, Raconte, Perspective AI, User Intuition and UserCall run interviews and return the transcripts, while Tally and Survicate create forms and surveys. To read research you already have, Dovetail searches a research repository, Userbrain reads usability tests, Intercom reads support conversations and Enterpret queries feedback from more than 50 channels.

### What is the difference between an MCP server and an API?

An API is written for developers: someone reads the documentation and writes code that calls each endpoint. An MCP server is written for an AI assistant: it describes each tool in plain language, and the assistant decides which one to call and with which parameters. Many products offer both, often with the same key. Raconte’s REST API and MCP server, for example, reach the same interviews and invitations.

### Are MCP servers free?

The connection is free on almost every server here. You pay for the product behind it. Tally’s server is free even on its free plan, Dovetail has a free plan, and Raconte includes 60 interview minutes each month at no cost. Perspective AI, User Intuition and UserCall give a one-off free allowance at signup. Survicate, Intercom and Userbrain start with a trial, while Enterpret is on quote.

### Can an AI agent interview users on its own?

Yes, through a server that runs the conversation. The agent creates the interview from a prompt, sends the link to the people it should reach, then reads the transcripts once they have answered. Raconte, Perspective AI and User Intuition support the whole loop. UserCall leaves sharing the link to a person on purpose. A person should still read the interview prompt before it goes out, because an agent tends to ask about what it already believes matters.

### Does ChatGPT work with MCP servers?

Yes. ChatGPT connects to remote MCP servers as connectors, and most vendors here list it as a supported client, including Perspective AI, User Intuition, UserCall, Tally, Survicate, Dovetail, Userbrain and Enterpret. Clients that offer one-click setup usually expect OAuth. A server that takes an API key, such as Raconte, needs the key pasted in the client configuration instead.

### Which MCP servers can read interview transcripts?

Raconte returns every transcript of an interview in one call, with a sentiment on each answer. Perspective AI returns conversations word for word, User Intuition opens transcripts and recordings, and UserCall returns transcripts when the full results are requested. Among reading servers, Dovetail reads the transcripts stored in its repository.

### Do I need OAuth or an API key to connect an MCP server?

It depends on the server. Survicate and Userbrain accept OAuth only, Raconte accepts an API key only, and most others accept both. OAuth suits a person working in Claude or ChatGPT, since each user signs in with their own account. An API key suits an agent that runs on its own, in a script, a scheduled job or an n8n workflow.

## Where to start

Reading feedback and asking for it are different jobs. A repository or a support inbox shows your assistant what users have already said. When your assistant needs to know why something happens, someone has to put the question to the users who go through it.

That is the job we built [Raconte](/en) for. [Connect the MCP server](/en/guides/mcp-setup) to your client with an API key, ask the assistant to create an interview and invite five customers, and read the transcripts with it the next day. You can start from [example requests for each tool](/en/guides/mcp-usage). You get the first 60 minutes each month free, then pay [by the minute](/en/pricing).

Ready to launch your first interview?

Describe your interview, share the link: an AI agent runs it by voice with each participant. You get back the transcript and analysis.

[Try for free](/login)

## Continue reading

[![Best user interview platforms 2026: 13 tools compared](/.netlify/images?url=_astro%2Fthumbnail.CKayV9Va.jpg&w=1376&h=768&dpl=6b031bf5959953076155d961)

August 1, 2026

## Best user interview platforms 2026: 13 tools compared

Most user interview platforms never ask a question. We compared 13 tools on who runs the interview, what it costs and whether you can call it from code.



](/en/blog/top-user-interview-platforms)[![Best customer feedback tools 2026: 11 tools compared](/.netlify/images?url=_astro%2Fthumbnail.BEBgL6FX.jpg&w=1376&h=768&dpl=6b031bf5959953076155d961)

October 4, 2026

## Best customer feedback tools 2026: 11 tools compared

We compared 11 customer feedback tools across interviews, surveys, in-app prompts and VoC platforms: who asks the follow-up, what it costs, how to plug it in.



](/en/blog/top-customer-feedback-tools)[![Context engineering: the context your AI agent is missing](/.netlify/images?url=_astro%2Fthumbnail.D2V7CmIa.jpg&w=1376&h=768&dpl=6b031bf5959953076155d961)

October 14, 2026

## Context engineering: the context your AI agent is missing

An AI agent is only as good as its context, which almost always lacks what users say. We show how to collect it and wire it in.



](/en/blog/context-engineering-ai-agents)
