SOAP2Docs ← Website
SOAP2 top banner # SOAP2 AI Assistant (Claude) — User Guide (BETA)

▶ Video AI Bridge example (0:42) · all videos

You can ask an AI assistant such as **Claude** many questions about SOAP2 in plain language and get answers taken from SOAP2's own Help, with links to the pages and feature guides they came from. Ask *"How do I rewind to a past date?"*, *"What is skew?"* or *"What changed in the latest version?"* and Claude looks it up in this documentation instead of guessing. There is no need to type 'run_analysis', '`search_help` or `get_help_page` — just ask naturally. NB: Claude is used here as an example. The SOAP MCP is a standard interface that can be used with any LLM platform that supports the MCP protocol. All major platforms do. Protocol: SOAP2.Mcp is a standard stdio MCP server. It forwards each call to SOAP2's local named pipe and returns JSON, and it never talks to a model. Any MCP-capable client can in principle launch it and call the same tools, and Gemini CLI and several other clients support MCP. - Server logic: the offline fallback, the confirmation check, the payload trimming and the Help search all live in the proxy and SOAP2. The model doesn't affect them, and the numbers returned are identical whichever AI asks. If a topic is advanced — the Greeks, implied volatility, skew, GEX/VEX — Claude is set up to start with a short plain-language introduction, then explain how SOAP2 handles it, then point you to further reading (SOAP2's own guides first, then well-known outside references, clearly labelled as outside SOAP2). **Premium** edition (or any higher edition, or the **Free Trial**) also unlocks Claude reading your actual SOAP2 data — trades, positions, vol/skew, GEX/VEX, regime tools — live while SOAP2 is running, or the last-stored numbers when it's closed. See Step 3. **Related:** - [SOAP2 Help Guide](./index.md.html) - [User Manual](./user_manual.md.html) --- # What you need - **SOAP2 installed.** Help search and live-state questions need SOAP2 *running* (nothing else has to be loaded — no Interactive Brokers connection, no open Options Analysis window). With **Premium or the Free Trial**, offline mode (Step 3) lets the AI Assistant still answer from SOAP2's last stored data even while it's closed. - **An AI assistant that can run local tools ("MCP servers")**, signed in: **Claude Desktop** or **Claude Code** (Anthropic), **Gemini CLI** (Google) or **Codex** (OpenAI, command line or IDE extension). Any other MCP-capable assistant can be connected too (see *Other AI assistants* below). Claude is the one we test with. - **Windows PowerShell** (already part of Windows) — used once to connect the assistant to SOAP2. Notes: - **SOAP2's connector is tested mostly with Claude.** Gemini CLI and Codex have their own one-click connect (Step 2) but have had less testing; tell us what you find. - **There is **nothing to switch on inside SOAP2**: the assistant's Help connection starts by itself every time SOAP2 starts. You only have to tell your AI assistant where to find it, once. # Step 1 — Check that SOAP2 is ready Start SOAP2. Open the **Message Log**; near the start you should see a line like: AI bridge started (local, current user only). Calls: …search_help, get_help_page That is all SOAP2 needs to do. If you don't see it, see *If something doesn't work* below. Notes: - **SOAP2 must be running** for live-state questions (Premium or Free Trial). - **The AI Assistant has a offline mode** (Premium or Free Trial) that reads the last-stored results from disk, so it can answer even when SOAP2 is closed. It is off by default; turn it on in **General Settings → Enable AI bridge (live state, beta)**. - **However, the initial SOAP2 AI setup cannot be skipped.** - **For Help search, SOAP2 can be closed** — Claude will read the last-stored Help pages from disk. # Step 2 — Connect your AI assistant (one time) In SOAP2, open **General Settings** and find **AI assistant**. Pick your assistant in the list (**Claude Desktop**, **Claude Code**, **Gemini CLI**, **Codex (OpenAI)** or **Other**) and click **Connect...** — there is nothing to open, find or type yourself. SOAP2 runs the connection for you and opens a small window showing what it did (leave that window open until it finishes; it tells you if Claude Desktop needs to be quit first). This is the same thing as running the script below by hand — the button just does it for you. **What "tested" means here.** Only **Claude Desktop** and **Claude Code** have been tested by the author. **Gemini CLI**, **Codex** and **Other** are marked **EXPERIMENTAL** in the list: when you select one, a line under the list says so and gives its specific warnings (what it needs and which file it changes), and clicking **Connect...** asks you to confirm before anything is changed. The numbers always come from SOAP2 and are the same whichever assistant asks; what can differ is how well the assistant follows SOAP2's instructions (for example asking before a large report, or the short answer format), and some assistants reject some tool definitions. If you try an experimental one, please tell us what happened on the Slack community, with your `MessageLog.txt`. Every connection can be undone: the scripts take `-Remove`. ## Claude Desktop 1. In General Settings, choose **Claude Desktop** and click **Connect...**. 2. If Claude Desktop is running, the window will tell you to quit it completely first — right-click its icon in the system tray (bottom right, near the clock) and choose **Quit**. Closing the window is not enough — Claude Desktop rewrites its settings when it exits, which would undo the change. Click the button again once it's quit. 3. The window prints the settings file it changed, the backup it made (a `.bak` file beside it), and *Done*. 4. Start Claude Desktop and open a **new chat**. SOAP2 appears among Claude's tools/connectors (look under the **+** or tools icon next to the message box). ## Claude Code 1. In General Settings, choose **Claude Code** and click **Connect...**. 2. Start a **new** Claude Code session. Typing `/mcp` in it lists **soap2** as connected. ## Gemini CLI 1. Install Gemini CLI if you have not (`npm install -g @google/gemini-cli`, then sign in the first time you run `gemini`). 2. In General Settings, choose **Gemini CLI** and click **Connect...**. The window adds a `soap2` entry to `.gemini\settings.json` in your user profile, keeps your other settings, and writes a `.bak` backup beside it. 3. Start a **new** Gemini CLI session. Typing `/mcp` lists **soap2**. ## Codex (OpenAI) 1. Install the Codex command-line tool (the Codex IDE extension shares its settings), so that the `codex` command works in PowerShell. 2. In General Settings, choose **Codex (OpenAI)** and click **Connect...**. The window runs `codex mcp add` for you (it writes `.codex\config.toml` in your user profile). 3. Start a **new** Codex session. Typing `/mcp` lists **soap2**. **About ChatGPT.** At the time of writing the ChatGPT app itself does not run local tools such as SOAP2's connector (it only takes tools that live on the internet), so it cannot be connected. Codex is OpenAI's assistant that can. ### If you'd rather do it by hand (or the button can't find PowerShell) SOAP2 installs a small connector program, **SOAP2.Mcp.exe**, together with its connection scripts (one per assistant), in the **AiAssistant** folder inside your SOAP2 installation folder. Open that folder, click in its address bar, type `powershell` and press Enter, then run: powershell -ExecutionPolicy Bypass -File .\Install-ClaudeDesktop.ps1 (or `.\Install-ClaudeCode.ps1`, `.\Install-GeminiCli.ps1` or `.\Install-CodexCli.ps1`). The `-ExecutionPolicy Bypass` part only applies to this one command; it is needed because Windows blocks downloaded scripts by default. Each script only edits its own assistant's settings. They are safe to run again, and all take `-Remove` to undo the connection: powershell -ExecutionPolicy Bypass -File .\Install-ClaudeDesktop.ps1 -Remove ## Other AI assistants (Cursor, VS Code and others — experimental) SOAP2's connector speaks **MCP** (Model Context Protocol), an open standard, so it is not tied to one company: any AI assistant that can run a local MCP server can use it. For an assistant that is not in the list, choose **Other (copy the MCP setting)** and click **Connect...**: SOAP2 puts the setting on the clipboard (a JSON form and a TOML form, with the path to **SOAP2.Mcp.exe** already filled in) for you to paste into that assistant's MCP settings. **We have only tested Claude properly**, so treat this as a do-it-yourself option. **How to connect one.** In your assistant's MCP settings, add a local (stdio) server named `soap2` that runs **SOAP2.Mcp.exe** from the **AiAssistant** folder, with no arguments. Most assistants take a JSON entry like this (use double backslashes in the path, and keep any servers you already have): { "mcpServers": { "soap2": { "command": "C:\\path\\to\\SOAP2\\AiAssistant\\SOAP2.Mcp.exe" } } } Check your assistant's own documentation for where its MCP settings live and for the exact format. Some assistants want extra fields (for example `"type": "stdio"`) or a different top-level name than `mcpServers`; the part that never changes is the command: the full path to **SOAP2.Mcp.exe**, no arguments. **Using the two scripts as a template.** The Claude scripts in the **AiAssistant** folder show exactly what is registered, and you can read them (they are plain text) before adapting them: - `Install-ClaudeDesktop.ps1` adds `mcpServers.soap2 = { "command": " \SOAP2.Mcp.exe" }` to Claude Desktop's JSON settings file, keeping every other entry. For an assistant that uses a JSON settings file, copy the script and change only the settings-file path. - `Install-ClaudeCode.ps1` runs the equivalent of `claude mcp add --scope user soap2 -- " \SOAP2.Mcp.exe"`. For an assistant with a command-line "add MCP server" command, the same shape applies: a name (`soap2`), then the exe path. Do not run the Claude scripts for another assistant as they are: they only edit Claude's own settings. Then start a **new** session and ask it to list its tools — **soap2** should be there. SOAP2 must be running for the live calls; Help search and the offline read work with it closed. **What may behave differently.** The numbers are the same whichever assistant asks, because they come from SOAP2, not from the AI. What can differ is how well the assistant follows the instructions that come with them: - **Asking before a comprehensive report.** It is an instruction to the assistant, not a lock. An assistant that ignores it may run the large report without asking. - **The succinct format** (header line plus exactly 5 one-line insights) and the offer of outside sources may be worded or laid out differently. - **Outside sources** come from the assistant's own web search, so their quality and format depend on it. - Some wording in this guide ("your AI allowance", "Claude") describes Claude and will not match another assistant exactly. If something does not work, the **Checking the MCP connection** section at the end applies to any assistant; just look in your assistant's own log instead of Claude's. We would like to hear what you find. # Step 3 — Live SOAP2 data (Premium, and the Free Trial) Everything above (Help search) works for every edition, with SOAP2 running or not. **Reading your live SOAP2 state** — open Options Analysis windows, trades, vol/skew/GEX readings, sticky-skew and dispersion regimes, watchlist — is a separate, still-in-beta capability included with the **Premium** edition, the **Free Trial** (like every other Premium feature during the test period), and always on for developers. It is read-only: nothing can be changed, no order can be placed, and no account balances are ever exposed. If you're on Premium or the Free Trial, turn it on in **General Settings → Enable AI bridge (live state, beta)**. Once on, ask things like: - *What's my current SPX position look like?* - *Give me a market read on SPX.* - *What's the sticky-skew regime right now?* - *Review Trade 25 for me.* - *Run SPX Distribution Stats and tell me what it shows.* Claude combines SOAP2's own numbers with outside information (general market knowledge, current events) to give you an actual read, not just a data dump — and it always tells you how old a figure is if SOAP2 hasn't refreshed it recently, rather than presenting it as live. ## Offline mode — works even with SOAP2 closed When SOAP2 isn't running (or the live-state setting above is off), Claude can still read SOAP2's **last stored results** straight off disk — the same numbers you'd see in each tool, just not refreshed. It combines those with outside information to give you an initial, clearly-dated read, then tells you which specific SOAP2 tool to run once you start it up, if something is stale or missing. This is on by default (**General Settings → Allow AI Assistant offline/outside-info access**) for the same editions as live state above. Turn it off if you'd rather Claude only ever answer from Help. ## Market analysis reports — `run_analysis` SOAP2 can write you a market read from its own calculations (SPX distribution stats, dispersion regime, sticky-skew regime and VIX term structure) at four levels of depth. Just use the **keyword** in your request: | Say… | You get | Cost | |---|---|---| | **"succinct"** — *"Run a succinct analysis."* | A bullet-point read: a header line (bar size and data date) and the **5 most important insights**, one line each. No trade idea. | Small | | **"comprehensive, size 250"** | Executive summary + one headline trade idea | Larger | | **"comprehensive"** (or *size 500*, the default) | Standard narrative connecting the findings across domains | Large | | **"comprehensive, size 1000"** (or *deep*) | Deep analysis with alternative scenarios and historical context | Largest | Start with **succinct** when you want a quick read; go to **comprehensive** when you want the full story. Both work when SOAP2 is running and when it is closed (offline, from the last stored results). **Claude asks before a comprehensive run.** It is long and uses a lot of your AI allowance, so Claude will stop and ask you first, mention that **succinct** is the cheaper option, and only run the comprehensive report after you say yes. Succinct runs straight away. **Outside sources are your choice.** The report itself only uses SOAP2's own data. By default Claude gives you the SOAP2 read first and then **asks whether you would like outside sources added** (for example today's economic calendar and Fed speakers, index and VIX news, earnings of the largest index weights). You can answer in advance instead: - *"Succinct analysis with outside sources."* — adds a clearly labelled outside-context section right away. - *"Succinct analysis, SOAP2 data only."* — no outside information and no question. Outside information comes from web searches Claude or another AI Assistant makes itself; it is not verified by SOAP2, so treat it as context, not as fact, and check anything that matters. **Comprehensive in more detail.** Ask Claude to **"run a comprehensive market analysis"** (with optional size 250/500/1000). It gathers the four sources above into one coherent AI-written summary with trade ideas. **Depth levels for comprehensive:** - **size 250**: Executive summary + one headline trade idea (for quick briefings) - **size 500**: Standard narrative connecting findings across domains (default; best for most use) - **size 1000+**: Deep analysis with alternative scenarios and historical context (for deep research) **Daily or hourly — never both.** The analysis uses one bar size throughout: **daily** (the default) or **hourly** (regular trading hours). It applies to the SPX distribution stats, the Sticky Skew read and the Dispersion read together, so the numbers Claude compares always come from the same kind of bar. Ask for it in plain words — *"Run a comprehensive analysis on hourly bars."* The hourly read only has something to work with once you have run the **Index Stats** and **Dispersion** tools in hourly mode at least once; the daily read needs them run in daily mode. Examples: - *"Run a comprehensive analysis."* (uses default 500 words, daily bars) - *"Run comprehensive analysis, size 250, just give me the headline."* - *"Deep comprehensive analysis please, 1000 words."* - *"Run a comprehensive analysis, hourly."* - *"Run a succinct analysis."* / *"Succinct analysis on hourly bars, with outside sources."* **It can use a lot of your AI allowance.** The comprehensive analysis hands Claude a large amount of data, so it uses noticeably more of your AI credits and conversation space than an ordinary question. That is why Claude asks first. Use succinct or size 250 when you only need the headline, and avoid re-running it several times in a row. **Online vs offline:** When SOAP2 is running, `run_analysis` gathers live or freshly-stored data and tells Claude which sources are fresh and which are stale. When SOAP2 is closed, it reads the last-known stored state from disk, clearly labelled as a snapshot. Either way, it returns sources + instructions for Claude to write the narrative. **What the offline snapshot contains.** To keep the amount of data manageable, the offline read sends one result per tool (two at size 1000, so Claude can compare them), shortens long histories to the most recent bars, and never includes your individual trades — ask about a specific trade separately. For tools that keep one result per option expiry, it uses the **nearest expiry that has not yet passed**. Every source is dated and marked **stale** if it is more than a day old, and anything left out is listed with the reason, so Claude can tell you what it did not have. # Understanding the AI Bridge calls Claude can use several different calls to gather information from SOAP2. Most of the time you don't need to think about which call Claude uses — just ask your question naturally. Claude automatically decides which tool fits your question best. But understanding how they work helps you ask for exactly what you want, and explains why Claude might suggest running a specific tool. ## How Claude chooses which tool to use You ask in natural language — *"What's my SPX position look like?"* or *"Run comprehensive analysis"* — and Claude sees that it has several SOAP2 tools available (via MCP, the Model Context Protocol). Claude reads your question, considers what information would help, and picks the right tool(s). You never type `run_analysis` or `get_market_read` directly. Those are the **tool names** Claude uses behind the scenes. When Claude runs a tool, it displays the result in your chat — you see the answer, not the tool name. If Claude picks the wrong tool or misses something, just clarify: *"Actually, I want the market read on SPX, not comprehensive analysis"* or *"Also run the distribution stats for me."* ## Help search (available in every edition) **`search_help` / `get_help_page`** - Search SOAP2's Help documentation by keyword or ask about a specific topic - Works offline; does not need SOAP2 running or live-state enabled - Includes SOAP2's guides, feature walkthroughs, and release notes - Examples: *"How do I rewind?"*, *"What's the difference between HSkew and VSkew?"* ## Live data and regime tools (Premium, Free Trial, and developers) These calls pull from open SOAP2 windows and stored analysis results. They work best with SOAP2 running, but fall back to offline mode (last-stored data) when SOAP2 is closed. **`get_market_read`** — SPX volatility, skew, and Greeks - Spot price, IV rank/percentile, IV curve, skew wings, GEX/VEX - Combines recent trade data (realized vol, realized skew) with current market - Use when you want a quick vol/skew snapshot: *"Give me a market read on SPX."* **`get_sticky_skew`** — Sticky-skew regime classification - Is the market in sticky-strike or sticky-delta regime? (and confidence level) - Daily or hourly breakdown - Use when regime is relevant to your decision: *"What's the sticky-skew regime right now?"* **`get_skew_read`** — HSkew and VSkew details - Horizontal skew (wings structure across strikes) - Vertical skew (term structure of skew across expirations) - Latest vs Previous comparison (what changed since you last looked) - Use for detailed skew analysis: *"What's the skew doing?"*, *"Has the wings structure changed?"* **`get_trade_review`** — Single trade P&L, Greeks, risk - For a specific trade number: Greeks, realized vs unrealized P&L, breakevens, legs summary - Works with trades in a loaded TLog or an open Options Analysis window - Use when you want a detailed take on one position: *"Review Trade 25 for me."* **`get_trade_rules`** — Trade alert rules - Per-trade rules monitoring (Time, Greeks, Financial, Positioning rules) - Status and recommended actions (Alert / Adjust / Roll / Reduce / Close) - Use to check rule alerts: *"Any trade rule alerts right now?"* **`get_cboe_gex`** — SPX, RUT or NDX gamma and vanna exposure from the CBOE delayed chain - GEX/VEX built from the **whole CBOE delayed option chain** (not from your own live IB data), for six scopes: 0DTE, 1DTE, end of month, monthly opex, opex+1 and aggregate. Gamma flip, walls and per-strike detail. - Each call fetches a fresh snapshot by default and compares it with the previous run when that run is at least 30 minutes old. The answer states the snapshot time and **"data as of"** (the newest trade time inside it), which is the number to judge freshness by, especially outside market hours. - Needs the **Premium GEX/VEX** licence and the live-state setting above. RUT and NDX are new; SPX is the default. - Use for: *"What does the CBOE GEX look like for 0DTE?"*, *"Compare RUT and NDX gamma walls."* **`run_analysis`** — Comprehensive market analysis and on-demand tool runs - **`run_analysis(comprehensive, size=250|500|1000, bar_size=daily|hourly)`** gathers SPX stats, dispersion, sticky-skew, VIX term into one AI-written summary with trade ideas, on daily (default) or hourly bars. Works online and offline. Large payload: it uses a significant amount of your AI allowance, so Claude asks you to confirm first (`confirmed=true`). - **`run_analysis(succinct, bar_size=daily|hourly)`** the same sources trimmed to a small payload, answered as the 5 most important insights. No confirmation needed; works online and offline. - **`outside_info`** (comprehensive and succinct): unset = SOAP2 data first, then Claude offers outside sources; `true` = add them now; `false` = SOAP2 data only. - **`run_analysis(spx_distribution_stats)`** starts a fresh SPX stats run (GARCH fit, tail risk, EVT analysis) - **`run_analysis(dispersion_regime)`** starts a fresh dispersion calculation and backfill - **`run_analysis(sticky_skew)`** starts a fresh sticky-skew calculation (daily, hourly, or hourly_eth) - Use for comprehensive briefings: *"Run comprehensive analysis."*, or to freshen a specific tool: *"Run spx_distribution_stats and tell me what it shows."* ## How they overlap Several calls gather related information: - **`get_market_read`** includes SPX stats (IV rank, realized vol, kurtosis) — these come from `run_analysis(spx_distribution_stats)`. If stats are stale, Claude will suggest running it. - **`get_market_read`** includes dispersion regime — the same data as `run_analysis(dispersion_regime)`. - **`get_sticky_skew`** returns the sticky-skew classification — the same calculation as `run_analysis(sticky_skew)`, but formatted differently (one structured answer vs detailed history). - **`run_analysis(comprehensive)`** combines data from all of the above (SPX stats, dispersion, sticky-skew, VIX term) into one coherent narrative, so you don't have to ask for each one separately. **Rule of thumb:** Use the specific calls (`get_market_read`, `get_sticky_skew`) for quick targeted questions. Use `run_analysis(comprehensive)` when you want to understand the whole market picture at once. # Step 4 — Ask questions Open a new chat and ask about SOAP2. Some examples: **Help and documentation:** - *How do I rewind to a past date in SOAP2?* - *What is skew, and what do the HSkew and VSkew charts show?* - *What are GEX and VEX? Is there a guide?* - *What keyboard shortcuts does SOAP2 have?* - *What changed in the latest release?* - *How do I set up email for sending feedback?* **Live data and regime analysis (Premium, Free Trial, developers):** - *Give me a market read on SPX.* - *What's the sticky-skew regime right now?* - *Review Trade 25 for me.* - *What's the skew doing? Has it changed?* - *Run comprehensive analysis.* (or *"Run it with size 250 for a quick summary."*) - *Run SPX Distribution Stats and tell me what it shows.* - *Any trade rule alerts?* You don't have to say "search the SOAP2 help" — Claude decides to look things up when the question is about SOAP2. If it doesn't, add *"check the SOAP2 help"* to your question. ## What a good answer looks like - It is based on SOAP2's documentation, and it says so if the Help does not cover something. (If you ask about something SOAP2 doesn't do, the right answer is "the Help doesn't cover that", not an invention.) - It **links to the sources**. Links go to soap2.dev. A page that exists only inside the installed SOAP2 is named instead of linked — open it from SOAP2's **Help** menu. - It mentions the relevant **feature guide** (the downloadable Word documents) when there is one. - For advanced notions it gives a short plain-language introduction first, and any outside reading is labelled as outside SOAP2. Claude can still make mistakes. For anything that affects a trade, check the linked SOAP2 page. # Keeping it working - **Updating SOAP2**: quit Claude Desktop/Code first if it's connected — the connector program (`SOAP2.Mcp.exe`) keeps running independently of SOAP2 itself for as long as Claude has it open, and an update installer can fail to replace it while it's locked. Re-running the connect scripts isn't needed unless the update notes say the connector changed; just restart SOAP2 (and Claude) after installing. - **Moving or reinstalling SOAP2** to a different folder: run the script again from the new **AiAssistant** folder so Claude points at the new location. - Help search and live-state questions only work **while SOAP2 is running**; without Premium's offline mode, Claude will say SOAP2 could not be reached if it's closed — start SOAP2 and ask again. # If something doesn't work | Symptom | What to do | |---|---| | No "AI bridge started" line in the Message Log | Restart SOAP2. If it still doesn't appear, another copy of SOAP2 may already be running for your Windows user — close it. | | *Could not find the AiAssistant folder* (from the Connect button) | Reinstall SOAP2, or use the by-hand steps above if you know where SOAP2.Mcp.exe is. | | The window says *Claude Desktop is running* | Quit Claude Desktop from the tray icon (see Step 2), then click **Connect Claude Desktop...** again. | | The by-hand script is blocked ("running scripts is disabled") | Use the full command shown above, starting with `powershell -ExecutionPolicy Bypass -File`. | | *SOAP2.Mcp.exe not found* | Use the **Connect** buttons in General Settings instead of a copied script, or run the by-hand script from inside the **AiAssistant** folder itself. | | SOAP2 doesn't appear in Claude | Make sure you quit and restarted Claude Desktop (or started a *new* Claude Code session) and opened a *new* chat. | | Claude says SOAP2 couldn't be reached | SOAP2 is not running, or was started after Claude's chat began. Start SOAP2, then start a new chat. | | Claude answers from general knowledge instead of the Help | Ask again with *"check the SOAP2 help"*. | | **MCP server connection lost** (Claude says "soap2 server failed to connect") | The SOAP2.Mcp process may have crashed, been stopped, or SOAP2 was rebuilt/updated. **Restart Claude Desktop or Claude Code** to reconnect. If SOAP2 was updated, restart SOAP2 first, then restart Claude. | | **SOAP2.Mcp won't start or is already running** | If SOAP2.Mcp gets locked (e.g., during a rebuild), close it: open Task Manager, find `SOAP2.Mcp`, right-click and **End task**. Then restart Claude. | | **MCP tools work, but then stop working mid-conversation** | A process (SOAP2 or SOAP2.Mcp) may have been restarted or updated behind the scenes. Start a **new chat** to reconnect. | ## Checking the MCP connection Claude Desktop keeps a log of its connections in `%LOCALAPPDATA%\Claude\Logs` — the file `mcp-server-soap2.log`, and `main.log` (search it for *soap2*). A line such as *Connected to soap2 (5 tools)* means the connection works. A line like *"Connection to soap2 failed"* or *"soap2 server disconnected"* means the MCP server (SOAP2.Mcp.exe) is not running, crashed, or SOAP2 was restarted. Restart Claude or the affected process. Include these logs when you [send feedback](./user_manual.md.html), together with SOAP2's `MessageLog.txt`. # Privacy and safety - **Everything is local.** The connection between Claude's connector and SOAP2 is a private Windows channel that only your own Windows user account can open. It is never reachable over the network. - **Read-only.** The assistant can only *read documentation*. It cannot place orders, change settings, open trades or touch your account. - **Help search alone** (every edition): Claude receives only the text of the Help sections it looked up — the same documentation published on soap2.dev — plus your question. It does **not** receive your trades, positions, balances or account numbers. - **With live-state/offline mode turned on** (Step 3, Premium or Free Trial, off by default except offline mode): Claude also receives the facts those calls describe — open trades' legs/Greeks/breakevens, vol/skew/ GEX readings, watchlist prices. It never receives account balances, net liquidation value, or IB account numbers (those are masked), and it is strictly read-only in every case — nothing can be changed, ordered, or transmitted. --- Go back to the main [SOAP2 Help Guide](index.md.html).