Product preview

Pressure-test a trade idea before your agent acts.

Connect Gryps to an MCP-capable AI assistant, then ask whether the market data is current, the expected move covers known fees, a venue fits the trade or a strategy behaves sensibly in a paper run. Gryps provides evidence and warnings. It does not trade.

See what you can askPreview the install
Package preview
Not published

Planned command

Planned npm command
npx -y @gryps/agent-core
Local processStdio transportNo wallet access
What you would do with it

Give your AI a trade idea. Ask it to check the practical details.

An AI assistant can summarise a signal or discuss a strategy on its own. Gryps gives it a controlled way to check the parts that are easy to overlook: which market the idea refers to, whether the data is available and fresh, what known fees may consume, and what the rules would do in a paper rehearsal.

Start with a trade idea

Paste in a signal, research note or question you already have. For example: “BTC may move 1.5%. Is that enough to matter?”

Add current market context

Gryps checks which market you mean, whether a current price is available, how fresh the source is and what known fees apply.

See the practical trade-offs

Ask whether the expected move still clears known costs, or compare which approved venue may better suit the asset and trade size.

Decide what to investigate next

Receive a sourced answer, a warning or a clear refusal. You can then discard the idea, investigate further or rehearse it in paper mode.

Questions you can ask

Four reasons to connect Gryps to your AI.

Start with a real decision you are already considering. Gryps adds market and execution context, explains what it found and stops before placing a trade. The technical tool names are included only as a reference for developers.

Use case 01Uses venue_snapshot + edge_check

Is this trade idea still attractive after known fees?

The situation
You receive a bullish BTC signal claiming a 1.5% move and want to know whether that headline survives basic execution costs.
Example prompt
Example prompt
“Check whether a 1.5% BTC signal still looks worthwhile after known fees. Show the market data you used and how fresh it is.”
What you get
A matched market, current price or unavailable reason, measured fee floor and a pass, warning or stop result with sources.
Boundary
This checks the maths and available market context. It does not prove that the 1.5% prediction is true.
Use case 02Uses venue_snapshot + route_compare

Which venue may better suit this trade?

The situation
You are considering a $25,000 BTC trade and want to understand how the approved venue options differ before choosing one.
Example prompt
Example prompt
“Compare approved venues for a $25,000 BTC trade. Explain the known fees, available market context and any reason not to recommend one.”
What you get
A ranked comparison with the reasons, source context and limitations behind the recommendation.
Boundary
Gryps can compare the options it is allowed to inspect. It cannot place or route the order.
Use case 03Uses signal_stack + edge_check

Do several signals really strengthen the case?

The situation
A research bot, a news model and an analyst all lean in the same direction, but they may be repeating the same underlying information.
Example prompt
Example prompt
“Combine these three BTC signals without treating related sources as independent. Show the effective case after known fees and flag overlap.”
What you get
A correlation-aware combined view, overlap warnings and an indication of whether the remaining case clears the measured fee floor.
Boundary
Gryps reduces obvious double-counting. It does not certify the quality or accuracy of the original sources.
Use case 04Uses paper_session

What would this strategy actually do?

The situation
You have a set of trading rules but do not want to learn how they behave with real money at risk.
Example prompt
Example prompt
“Run these rules in paper mode for 20 decision points. Show every action, every decline and the estimated cost of unnecessary activity.”
What you get
A narrated paper session showing what the agent would do, what it would reject and how much known friction the activity may create.
Boundary
The session is a rehearsal only. It does not connect a wallet, use an account or move funds.
How it connects

Useful checks without trading access.

Your AI sends a question through the local Gryps package. Gryps calls only approved read-only or paper tools, then returns evidence, warnings or a clear refusal. The connection has no wallet or order authority.

Your AI assistant

You ask a trading question in a client that supports local MCP tools.

Local Gryps connector

The package translates the question into one approved Gryps check.

Approved market sources

Gryps returns available market evidence, known costs or a clear refusal.

Install flow preview

Familiar to install. Honest about what is not ready.

The intended path is one local npm command, one client entry and one safe first question. The command remains visibly locked until the package name and registry publication are confirmed.

Step 01

Install the local server

Terminal command
npx -y @gryps/agent-core

Preview only. The package is not on npm and the final package name is still subject to release approval.

Step 02

Add Gryps to your MCP client

MCP client configuration
{
  "mcpServers": {
    "gryps": {
      "command": "npx",
      "args": ["-y", "@gryps/agent-core"]
    }
  }
}

The exact settings surface varies by client. The package is intended for clients that can launch a local stdio MCP server.

Step 03

Ask a question you can verify

Safe first prompt
“Check the current BTC market before we do anything else. Tell me which market you matched, the latest available price, how fresh the source is, and clearly say if anything is unavailable.”

A safe first response should prove identity and freshness before inviting further analysis. If price data is unavailable, the server should return a typed reason rather than substitute an estimate.

PrivacySecuritySupport and feedbackDecision support only. Not financial advice.
Public access boundary

Publicly useful does not mean broadly privileged.

The public profile is designed as a separate, allowlisted capability surface. It does not inherit the permissions of Gryps internal agents or expose private operating material.

Intended public profile

Read and rehearse

  • Approved venue and product data
  • Paper sessions and recommendations
  • Freshness, provenance and structured refusal
  • No transaction side effects

Explicitly excluded

No hidden authority

  • No order placement, signing or custody
  • No private repositories or internal calls
  • No arbitrary file or credential access
  • No direct path to production writes

Release gates

Verify before publish

  • Final package name and registry entry
  • Approved public tool catalogue
  • Authentication, quotas and audit controls
  • Externalisation and ownership clearance
Release posture

Preview now. Publish when the boundary is real.

The local server and tool architecture exist. Public release waits for the package, tool catalogue, access controls and ownership gates to be confirmed together.

Read the Gryps Agent Standard