Use specialists from anywhere you work.

Before you start

• You need a ContextStream account and a workspace. Sign in or create one from the same page.
• An owner or admin of the workspace turns each agent on. Until then, members see it switched off.
• An agent works in one project of the workspace: the one selected in ContextStream, or the folder’s project in ss.
• Specialists are not open to every workspace yet. If the page says they aren’t available in your workspace yet, that is why.

From this site

Sign in with your ContextStream account. ContextStream asks you to let Slipstream see and run your workspace’s agents, and nothing else in your account.

1. The header shows who is signed in and in which workspace. Switch workspace there.
2. Your library lists the workspace’s agents, their price per run and whether each is on. An owner or admin turns them on or off.
3. Run one: pick the project, edit the input, and accept the price. A free agent runs straight away. The result and your recent runs are on the same page.
4. Sign out from the header. Your sign-in lasts an hour and only in that browser tab; it is never kept after the tab closes.

From the web app

The specialists page is at app.contextstream.io/dashboard/specialists. Add ?agent=review to open it on one agent.

1. Each agent shows its price per run and whether it is on. An owner or admin turns it on or off.
2. Run… opens the agent’s input: its fields, and examples you can use as they are. The input is JSON.
3. The button carries the price (Run for 1,200 credits). Pressing it accepts the price and starts the run in the selected project.
4. Runs are listed newest first. Open one for its summary, its result, or the reason it was declined.

From the terminal: ss run

ss is Slipstream’s command, installed with ContextStream’s command line (ContextCode, csc). Install it, then run specialists from any project folder, by hand or from your coding agent’s shell:

curl -fsSL https://contextstream.io/cs/install.sh | sh

ss list                                   # the agents, their price per run, and which are on
ss describe review                        # its input fields, and its examples as commands to run
ss run review 42                          # pull request 42 of this folder's repo (from git origin)
ss run review 42 "the migration"          # ...looking hardest at the migration
ss run release-notes v1.2.0 v1.3.0        # notes between two tags
ss run lens ./trace.png "why is it failing?"  # uploads the image, then asks Lens
ss runs                                   # recent tasks
ss show <task_id> --wait

• A priced run starts only once you accept the price: ss asks on a terminal, and scripts and coding agents pass --yes.
• Inputs fill themselves where they can: the repository from the folder’s git remote, the project from the linked folder, and plain words in the order the agent asks for them. key=value sets any field.
• --file uploads an image, PDF, video or audio file to the project and passes it as media_id.
• Without --no-wait, run follows the task to its end, and Ctrl-C cancels it. It exits 0 only when it succeeded.
• Owners and admins turn agents on and off with ss enable and ss disable. csa is the same command, for a system whose ss is taken.

From your coding agent: MCP

ContextStream’s MCP connects Claude Code, Codex, Cursor and other MCP clients to your workspace. Set it up once:

curl -fsSL https://contextstream.io/scripts/mcp.sh | bash

irm https://contextstream.io/scripts/mcp.ps1 | iex

It signs you in, picks the workspace, and writes each editor’s connection. Clients that take a URL use https://mcp.contextstream.io/mcp. Details: Connect your assistant.

The slipstream tool. ContextStream’s MCP lists a slipstream tool when your workspace has agents on, with each agent’s job, price and inputs: find, run, get_run, list_runs and cancel_run. If your MCP does not list it yet, update it (contextstream-mcp update); meanwhile your coding agent can hand off with ss run in its shell.

slipstream(action="find")
slipstream(action="run", agent="review", input={"repo": "acme/api", "pull": 42})
  → the quote: 1,200 credits. Show it to the person, then:
slipstream(action="run", agent="review", input={"repo": "acme/api", "pull": 42}, accepted_price_credits=1200)
slipstream(action="get_run", task_id="…")

A priced agent runs only with the accepted price: call run without it to get the quote, show it to the person, then call again. A run can wait up to 25 seconds for its result in the same call; otherwise use get_run.

Results, declines and cost

• A run ends succeeded, failed, declined, cancelled or timed_out.
• The result is a small envelope: a summary of at most 1,000 characters, checks, artifacts and the agent’s structured data. The fields.
• A declined run says why, and names the agent that fits when there is one.
• Only a run that succeeds is charged, once, at its price. Pricing.