API and MCPContext in your code and your AI tools
Put agents to work from your code
Start a task with one request and get the finished files back, or connect the AI tools you already use to your Context files, apps and skills.
Terminal
Run Context's agent in your terminal. Sessions appear here as tasks.
After installing, this prints a pairing code and opens the approval page. Approve it and the terminal is connected.
Machines that run dispatched tasks, yours in every workspace, whichever one you paired it from.
Start a task with one request
Send the brief to the task API and get a task back at once. Retries with the same key never start it twice.
Get the finished work back
Read the task when it is done: the answer, the files it made and the skills it used.
| Name | Written by | For | Modified | Size |
|---|---|---|---|---|
| q3-portfolio-review.pptx | AtlasAgent | Maya Okafor | 1h ago | 4.1 MB |
| task.md | Context | Maya Okafor | 1h ago | 3 KB |
Give each script its own key
Create a key with only the capabilities a script needs, limit it to chosen agents, and set it to expire.
New task key
For scripts that create tasks, upload files, or read evals.
Run agent tasks from any system
The task API is how a CRM, a data pipeline or a scheduled job hands work to an agent and collects the result.
Scope every key to its job
Admins create keys in Settings, each one limited to what the script that holds it has to do.
New task key
For scripts that create tasks, upload files, or read evals.
New task key
For scripts that create tasks, upload files, or read evals.
Reach Context from your AI tools
Context runs an MCP server at your deployment's address, so tools like Claude Code, Codex and Cursor can work with your files, apps and skills.
Secure by default
Every request carries a key or a token tied to a person or an agent, and the work it starts runs under the same rules as work started in the app.
Audit log
Who granted, shared and revoked access across the organization, and every agent permission decision, with who answered.
| Time (UTC) | Outcome | Event |
|---|---|---|
| Today | ||
| 09:39:02.118 | Allowed | Atlas changed q3-portfolio-review.pptx · approved by Priya Natarajan |
| 09:38:41.774 | Asked | Atlas asked to change q3-portfolio-review.pptx · sent to a person to decide |
| 09:38:40.210 | Recorded | Atlas read q3-valuation-bridge.xlsx · for Priya Natarajan |
| 09:31:12.530 | Recorded | Maya Okafor opened Covenant headroom |
| 09:31:05.064 | Granted | Ledger published Covenant headroom · to Brindle Capital |
| 09:30:52.301 | Recorded | Ledger created Covenant headroom · for Maya Okafor |
| 09:30:48.957 | Granted | Ledger connected Q3 portfolio review deck · to Covenant headroom, read only |
| 09:24:48.902 | Recorded | Atlas wrote q3-portfolio-review.pptx · under Q3 portfolio review deck, for Maya Okafor |
| 09:24:12.388 | Recorded | Atlas read q3-portfolio-updates.docx · for Maya Okafor |
| 09:21:30.217 | Granted | Maya Okafor granted Editor on Q3 portfolio review deck · to Ledger |
| 09:20:11.645 | Recorded | Maya Okafor sent a message in #q3-portfolio-review |
| 09:13:27.390 | Recorded | Ledger wrote q3-valuation-bridge.xlsx · under Q3 portfolio review deck, for Maya Okafor |
| 09:04:55.081 | Recorded | Scout wrote q3-portfolio-updates.docx · under Q3 portfolio review deck, for Maya Okafor |
| 09:02:19.446 | Recorded | Maya Okafor sent a message in #q3-portfolio-review |
Showing 1 to 14 of 97 events
Atlas (on behalf of Maya Okafor) changed q3-portfolio-review.pptx
- Requested by
- Atlas
- Decided by
- Priya Natarajan
- On behalf of
- Maya Okafor
- Acting mode
- On behalf of its user (agent_obo)
- When
- Today, 09:39:02 UTC
- Event type
- gated_action_approved
- Event ID
- aud_7c41e09b2f
Audit log
Who granted, shared and revoked access across the organization, and every agent permission decision, with who answered.
| Time (UTC) | Outcome | Event |
|---|---|---|
| Today | ||
| 09:39 | Allowed | Atlas changed q3-portfolio-review.pptx · approved by Priya Natarajan |
| 09:38 | Asked | Atlas asked to change q3-portfolio-review.pptx · sent to a person to decide |
| 09:38 | Recorded | Atlas read q3-valuation-bridge.xlsx · for Priya Natarajan |
| 09:31 | Recorded | Maya Okafor opened Covenant headroom |
| 09:31 | Granted | Ledger published Covenant headroom · to Brindle Capital |
| 09:30 | Recorded | Ledger created Covenant headroom · for Maya Okafor |
| 09:30 | Granted | Ledger connected Q3 portfolio review deck · to Covenant headroom, read only |
| 09:24 | Recorded | Atlas wrote q3-portfolio-review.pptx · under Q3 portfolio review deck, for Maya Okafor |
| 09:24 | Recorded | Atlas read q3-portfolio-updates.docx · for Maya Okafor |
| 09:21 | Granted | Maya Okafor granted Editor on Q3 portfolio review deck · to Ledger |
| 09:20 | Recorded | Maya Okafor sent a message in #q3-portfolio-review |
| 09:13 | Recorded | Ledger wrote q3-valuation-bridge.xlsx · under Q3 portfolio review deck, for Maya Okafor |
| 09:04 | Recorded | Scout wrote q3-portfolio-updates.docx · under Q3 portfolio review deck, for Maya Okafor |
| 09:02 | Recorded | Maya Okafor sent a message in #q3-portfolio-review |
Showing 1 to 14 of 97 events
Atlas (on behalf of Maya Okafor) changed q3-portfolio-review.pptx
- Requested by
- Atlas
- Decided by
- Priya Natarajan
- On behalf of
- Maya Okafor
- Acting mode
- On behalf of its user (agent_obo)
- When
- Today, 09:39:02 UTC
- Event type
- gated_action_approved
Use it wherever your code runs
Call the API from any language that speaks HTTP, use the command line from a terminal, and watch the same tasks in the web and phone apps.
Terminal
Run Context's agent in your terminal. Sessions appear here as tasks.
After installing, this prints a pairing code and opens the approval page. Approve it and the terminal is connected.
Machines that run dispatched tasks, yours in every workspace, whichever one you paired it from.
Work across the Context platform
A task started from code runs on the same engine, with the same skills, connectors and audit log, as one started in the app.
- SkillsName a skill in the request, and the task records every skill it applied.
- EvalsRead traces, scores and analytics for your agents through the read-only Evals API.
- ConnectorsTasks from the API reach the same apps and services your connectors offer.
- Audit logAgent actions and the decisions on them, whether the task started in the app or from code.
Home
- q3-portfolio-review.pptx1h ago
- q3-valuation-bridge.xlsx2h ago
- q3-portfolio-updates.docx3h ago
- lp-letter.pptx5h ago
Home
- q3-portfolio-review.pptx1h ago
- q3-valuation-bridge.xlsx2h ago
- q3-portfolio-updates.docx3h ago
- lp-letter.pptx5h ago
Deploy in your environment
Run the API and the MCP server inside your own deployment, on your own address, behind your own identity provider.
- Your cluster, your addressThe API and the MCP server run on your own Kubernetes cluster, at your deployment's address.
- Your model gatewayReach models through the Context Gateway, LiteLLM, Amazon Bedrock, Azure OpenAI or a PrivateLink endpoint.
- Air-gapped installationInstall without internet access, with the air-gapped installation tested in CI.
- Your identity providerSign people in with SAML or OIDC and provision them with SCIM.
Context runs in production at Qualcomm. Read the case study
- Amazon EKS
- Azure AKS
- Air-gapped
The API and MCP server are included in every Context workspace, including the Free plan.
- Free$0Includes API and MCP
- Plus$20Per person per month
- EnterpriseCustomEnterprise deployments run in your own cloud.Talk to us about a plan
Questions
What can the task API do?
Create a task, read its status and output, cancel it, share it and download the files it made. It also uploads files, manages workflows and reads evals.
What happens if my request is sent twice?
Each create carries an idempotency key. A repeat with the same key returns the task already started instead of starting another.
How do webhooks work?
Ask for webhook delivery and give a signing secret when you create the task. When it finishes, Context sends the result signed with that secret, with an event id and a timestamp, and tries up to three times.
Which AI tools work with the MCP server?
Any MCP client that supports OAuth sign-in, such as Claude Code, Codex and Cursor. Point it at your deployment's MCP address and approve it on the consent page.
Is there an SDK?
The API is plain HTTP with JSON, so any language can call it. Ask your account team for the API reference for your deployment.
Who can create keys?
Admins create task keys in Settings > API Keys. Each key has the capabilities you choose, an optional list of agents and an optional expiry.



