MANAGE & OPERATE
Context Graph API
Give your agents an authenticated, connected view of API and service relationships. Trace downstream impact, identify affected consumers, and catch breaking changes before you merge every line of code.
One graph, built from everywhere your APIs actually live
Connect what you already have
Your Postman workspaces, your GitHub repos, and your running services in New Relic. Connect once.
Every edge shows its work
Typed nodes, six typed edges, and evidence on each one: the commit it was seen at, and where the signal came from.
Answer impactful questions to accelerate your agentic SDLC
Blast radius, dependency chains, endpoint consumers, ownership.
How Context Graph works
Benchmarked across an estate of 468 repositories
Eight questions, three models, three runs of each arm. Same model, same questions, same tools, with and without the graph.
29%
fewer prompt tokens
Opus 4.7. GPT-6 Astra fell 23%. Sonnet 4.5 used more.
17%
fewer tool calls per run
Opus 4.7, 69 calls down to 57.
27%
lower cost per run
Opus 4.7. GPT-6 Astra fell 20%. Sonnet 4.5 rose.
September 2026 benchmarking: 468 repositories, three models, 144 runs. Cost excludes the one-time graph build and ongoing re-ingestion.
Fewer tokens on the questions that span systems
Reduction in prompt tokens when the model queries the graph before it reads any code, Opus 4.7 over 468 repositories. These are the three largest reductions of the eight questions we asked. The savings come from fewer round trips: 57 tool calls per question instead of 69.
Chart will load when visible
GETTING STARTED
Try Context Graph directly in Postman with a pre-built ask against the live API.
1. Context Graph is authenticated, so sign in to your Postman account before running queries.
2. Add a Postman API key in the request's Authorization tab.
3. Set the active environment to Context Graph API.
4. Each graph is populated per team, so whichever team you're signed in as is the team whose nodes get ingested and queried.
5. Edit the query in the request body to ask your own question.
You can also view your ingested graph any time from the Postman home button → Agent Context.
USE IT FROM YOUR CODING AGENT
Reach the same graph whichever way your agent works: from the CLI, through MCP, or over HTTP.
CLI + skills. Run postman init in your repo to install Postman's agent skills, then your coding agent queries the graph through the Postman CLI, with no HTTP wiring or polling to manage.
MCP. Point any MCP client at https://mcp.postman.com/context-graph, or add the --context-graph toolset to the Postman MCP server, and the graph shows up as a tool.
HTTP. Call the Context Graph API directly with a Postman API key: submit an ask and poll for the result.
See how the Context Graph helps with a multi-repo change
RESOURCES
Go deeper on Context Graph
Frequently asked questions
What is Postman Context Graph?
A continuously updated graph of your API estate. It ingests your Postman workspaces, code repositories, and runtime telemetry, then turns them into typed entities (APIs, endpoints, deployments, databases, teams) joined by typed relationships. Agents query that graph instead of reconstructing it from whatever files are open.
Which sources can Context Graph ingest?
Your Postman workspaces are the primary source: the specs, collections, and environments your teams already maintain. Beyond that, GitHub repositories scanned for OpenAPI, AsyncAPI, and service definitions, and New Relic for the deployments and telemetry a code scan cannot see.
How does the graph stay current?
Each source carries its own policy: a refresh interval, a webhook, or manual. Re-ingestion is incremental, so unchanged repositories are skipped and cost tracks your code churn rather than your estate size. Runtime connectors stream continuously.
What does Context Graph know about an API?
What it exposes, calls, and depends on at runtime. Which databases back it, which external services it talks to, which team owns it, where it is deployed, and what watches it. Every relationship carries its evidence, so you can tell a confirmed edge from an inferred one.
Can I use Context Graph through MCP?
Yes. Point any MCP client you already run (your editor, your agent framework, your own harness) at https://mcp.postman.com/context-graph, or add the --context-graph toolset to the Postman MCP server, and the graph is available as a tool, without writing the HTTP request or handling the asynchronous polling yourself. The HTTP endpoint is still available for anything you'd rather build against directly.
When should I not use the graph?
When the answer is already in the files in front of you. Our benchmarking found that on a declared dependency, like who imports a module, go.mod answers outright: reading the repository scores 98% on its own and the graph adds nothing. It earns its keep on cross-boundary work.