A trading agent is only as good as the data it reasons over. Simmer gives your agent first-class access to its core surface — markets, prices, positions, context, and briefing across every venue (see Building Skills). For everything beyond that — a data feed you subscribe to, a broker or analytics API you hold keys for, an internal tool you operate — you can bring your own data: generate a clean, agent-native interface for it and plug it into your skill.
Only bring data you are authorized to use. This pattern is for sources you already have rights to — your own API keys, subscriptions, accounts, or data. Many services prohibit automated access or redistribution in their Terms of Service; some prohibit reverse-engineering. You are responsible for ensuring your use is permitted. Simmer does not host, proxy, cache, or redistribute anything you generate this way — it runs entirely on your machine, under your own access and credentials.
The pattern
The SDK deliberately doesn’t bundle third-party API clients — it stays scoped to Simmer’s surface plus universal primitives. For a one-off call, hitting the source’s API directly from your skill is fine (see External market data). But when you want a reusable, agent-native interface to a source — one your agent can call thousands of times a day without burning tokens on hand-rolled request payloads — generate a dedicated CLI/MCP for it.
The tool for this is Printing Press (open-source, MIT). It turns an API into a token-efficient CLI and an MCP server, shaped for agents: typed exit codes, auto-JSON when piped, --compact output, a local SQLite cache. Three input modes:
- OpenAPI spec — if the source publishes one.
- HAR file — export your own authenticated session’s traffic from your browser’s DevTools.
- URL — for a source you operate or are entitled to script, it can capture traffic and reverse-engineer a spec.
Workflow
Install Printing Press
Requires Go 1.26.4+ and an agent that loads open-agent-skills (Claude Code is the tested path).curl -fsSL https://raw.githubusercontent.com/mvanhorn/cli-printing-press/main/scripts/install.sh | bash
Generate the interface for your source
Inside your agent, point it at the API by name, a spec, or a URL you’re authorized to use:/printing-press <your-source>
<source>-pp-cli (for shell agents) and <source>-pp-mcp (an MCP server) — from one spec, sharing the same client, store, and auth. Provide your credentials
The generated CLI reads auth from environment variables — your keys for your access. Run cli-printing-press auth doctor to confirm they’re set. Nothing is sent to Simmer.
Register the MCP with your agent
Add the generated <source>-pp-mcp server to your agent’s MCP config. Now your agent can call the new data source as a native tool — in the same loop where it reads Simmer context and places trades.
Using it in a skill
Once your agent can read the external source, the skill shape is the same as any other Simmer skill: turn the data into one measurable signal, one entry gate, one sizing rule, one exit rule (see Building Skills). Your agent fetches from your data source, forms a probability or signal, and executes through SimmerClient with sizing and safeguards:
from simmer_sdk import SimmerClient
from simmer_sdk.sizing import size_position
# Your agent reads your data source (via the generated MCP/CLI),
# turns it into a probability, then trades through Simmer:
amount = size_position(p_win=my_probability, market_price=price, bankroll=bankroll, min_ev=0.03)
if amount > 0:
client.trade(market_id=mid, side="yes", amount=amount,
source="sdk:my-skill", skill_slug="my-skill",
reasoning="Signal from my own data source diverges from market")
This is a developer-grade path. It needs Go, a skills-capable agent harness, and (for the HAR/URL modes) a manual capture step. If you’d rather not run it, hitting your source’s API directly from the skill (as in External market data) is the lighter option for simple cases. What Simmer touches
Nothing. The generated CLI/MCP runs on your machine, authenticates with your credentials, and feeds your agent. Simmer never sees, stores, or redistributes the external data — your agent simply arrives at its trades better-informed. That’s the whole point: the data relationship stays yours. 