Web Answer
One question in, one grounded, cited answer out. No model to choose, no retrieval pipeline to build.
Why Web Answer?
Answering one question with sources is more than a single API call. You call a search API, decide which results to trust, retrieve and clean the pages, prompt a model to synthesize, and build citation extraction so the output can be traced back to a source, often looping back to search again when the first pass does't fully answer the question.
Web Answer collapses that into a single call. Your agent sends a question and gets back a finished answer, grounded in current web sources, with citations that link the answer back to the sources it drew on.
It runs on Amazon Bedrock AgentCore as a managed connector on Gateway, reached over MCP.
Benefits
What changes when you get back a finished answer.
Today a grounded answer means calling a search API, picking which results to trust, retrieving and cleaning pages, prompting a model, and writing your own citation extraction. Every step is undifferentiated work you have to build and maintain. Web Answer takes one question and returns the finished answer. You choose no model, write no synthesis prompt, and maintain no search loop for that sub-task.
Every grounded answer comes back with its sources attached. Each citation names the source domain and links to the page where one is available, so every claim is traceable to its origin. Citations are returned when the answer draws on web sources.
Access is authorized with IAM, so there is no separate credential to provision or rotate, and Web Answer attaches to a Gateway you already run. Web Answer does not store your queries or results, and your data is never used to train models. Per-account rate limits and abuse guardrails are on from the first call, and operational metrics publish to CloudWatch.
Web Answer is billed per tool call, not per token, so the cost of each answer is the same whether it comes back short or long. You are not exposed to a long answer quietly costing more, which makes an agent turn easy to budget. Billing is $10 per 1,000 tool calls; see the Amazon Bedrock AgentCore pricing page for current rates.
Where teams use Web Answer
Three jobs where you want an answer, not ten links to read.
Answers a compliance reviewer can check
In banking, healthcare, and the public sector, a wrong answer carries real cost, so attribution matters. Every grounded answer comes back with its sources, and each statement links to the source it came from, so a reviewer can verify any claim at its origin.
Customer service, answers about the outside world, in real time
When a service agent is on a live contact, some questions can't be answered from internal content alone. They need current public requirements, third-party product details, travel or market conditions, anything that changes on the public web. Web Answer returns a grounded, cited response in a single call, giving the agent sourced material to inform their reply, with citations the team can review afterward. Responses return in seconds.
One question inside a larger agent
Often a question is just one step in a longer workflow: check a claim against current sources, see what changed in a market, confirm a figure before relying on it. Web Answer handles that step in a single tool call, so the agent you are building does not carry a retrieval pipeline it never wanted.
Resources
Frequently asked questions
Frequently asked questions
Open allThe dividing line is who runs the model. Web Search returns ranked results and you keep your own model and your own synthesis. Web Answer runs the model and the loop and hands back a finished answer. Use Web Search when you want control over how the answer is composed, and Web Answer when the question is a step inside a larger agent and you do not want to own a retrieval pipeline for it. They are complements, and many teams use both.
Web Answer is a managed tool exposed over MCP on an Amazon Bedrock AgentCore Gateway, so any agent attached to that Gateway can call it, whichever framework you build in. AWS runs the model behind the answer, so there is no model for you to select, size, or manage.
It is a tool your agent calls programmatically over MCP, not an end-user interface. The only console touchpoint is where a Gateway administrator adds Web Answer as a target and configures it.
A Gateway administrator adds Web Answer as a target and sets the configuration that carries cost and policy, including filter bounds and source settings. At runtime the model sees only the fields the administrator left open, such as the query. That split is also how you constrain what a call can cost, because the settings that drive cost sit in the target configuration rather than in the prompt.
Yes. Web Answer supports domain include and exclude lists, so you can restrict answers to sources you trust or rule out sources you do not want, and publish-date filters to bias toward fresher material. An administrator sets the bounds on the Gateway target, and the agent works within them.
English at launch, with more languages to follow. Web Answer launches in one Region, US East (N. Virginia), with more Regions to follow. See the Amazon Bedrock AgentCore supported Regions table for the current list.
No. Web Answer does not store your queries or results, and your data is never used to train models. Access is authorized with IAM under your own account, and traffic can egress through your VPC or PrivateLink rather than the public internet.
Web Answer is billed per tool call rather than per token, at $10 per 1,000 tool calls. That makes the cost of an agent turn forecastable per question rather than varying with the length of the answer. See the Amazon Bedrock AgentCore pricing page for current rates.
Get started
Add Web Answer to a Gateway and ask it a question.
Did you find what you were looking for today?
Let us know so we can improve the quality of the content on our pages