AI connectors
Connect your AI assistant to FSOtools.
An MCP connector is an address you give your AI assistant, such as Claude, so that it can look things up here while you talk to it. MCP (Model Context Protocol) is the common way assistants connect to outside sources. Once a connector is added, you keep asking questions in your own words; the assistant decides when to consult FSOtools, fetches the matching records and quotes them back with their official links. There is nothing to install, no account and no key.
Every connector here serves reference material only and stores nothing about you. Published decisions, directive text and reviewed guidance go out; nothing about you or your question is kept. The one exception is explicit: if you choose to offer an answer to the Knowledge Desk through the Question Bot connector, the name and e-mail you give for attribution are held for review.
The connectors
One entry per server FSOtools publishes on this domain, with its status. Live means the address answers today. Planned means it is designed or built but not deployed, so it has no address to add yet; a planned connector gains its connect command here on the day it goes live.
- Live
- Planned
Adjudication Atlas Live
Searches, compares and counts the published DOHA security-clearance decisions in the Adjudication Atlas: case number, date, guidelines, cited SEAD 4 conditions and outcome, the Atlas's statistics with their confidence intervals, and a search of the decisions' words that returns short passages, every result linked to the official decision on DOHA's site.
| Address | https://fsotools.ai/atlas/mcp (Streamable HTTP). Opened in a browser, the address shows a guide to its six tools. |
|---|---|
| Claude Code | claude mcp add --transport http doha-atlas https://fsotools.ai/atlas/mcp |
| Claude.ai | Settings → Connectors → Add custom connector. Name it Adjudication Atlas, paste the address and save. |
| Privacy | Stateless: it issues no session and keeps none. Request logging is switched off in the Worker, so no record of what was asked exists. The only limit is 60 requests a minute from one network address. It answers from published decisions only: no applicant details, no case narratives, and it predicts nothing. |
| Product page | Adjudication Atlas, served by the doha-atlas-mcp Worker on this domain. |
FSO Question Bot Live
Answers a Facility Security Officer's compliance question with cited passages from the approved DCSA library release (32 CFR Part 117, ISLs, DCSA and ODNI publications), a reviewed Knowledge Desk answer where one matches, and related published DOHA decisions. It never composes an answer itself: your assistant does, from the citations it returns.
| Address | https://fsotools.ai/question-bot/mcp (Streamable HTTP). Opened in a browser, the address shows a guide to its seven tools. |
|---|---|
| Claude Code | claude mcp add --transport http fso-question-bot https://fsotools.ai/question-bot/mcp |
| Claude.ai | Settings → Connectors → Add custom connector. Name it FSO Question Bot, paste the address and save. |
| Privacy | Stateless: it issues no session and keeps none, and request logging is switched off in the Worker. The limit is 60 requests a minute from one network address. Every result carries its citation (document, passage, official link), so nothing is quoted without a source. One tool is different by design: if you choose to offer a researched answer to the Knowledge Desk, the connector asks for your consent and the name and e-mail to credit, and holds that submission for the maintainer's review. Nothing is stored unless you say yes. |
| Product page | FSO Knowledge Desk, which publishes its accepted answers; the projects page lists the Question Bot itself. Served by the fso-question-bot-mcp Worker on this domain. |
Adverse Information Assistant Planned
Reference material for someone preparing an adverse-information self-report: which SEAD 3 reporting table applies, the text of the SEAD 4 guideline and conditions a report will be measured against, and the maintainer's checklists and form maps. The report itself is prepared in your own assistant, which never has to send the server a word of your situation.
| Status | Designed; not built, not deployed. Its planned address is https://fsotools.ai/adverse/mcp. |
|---|---|
| Privacy | By design, stricter than the Atlas connector: stateless, no request logs, no sessions, and no free text at all. Every request is a code from a published list (a guideline letter, an access level, an event type), so there is nothing about your matter that the server could receive, let alone keep. |
| Product page | Adverse Information Assistant on GitHub ↗ (open source). |
How to connect
The steps are the same for every live connector. Only the name and address change.
| Claude.ai | Settings → Connectors → Add custom connector. Give it the connector's name, paste its address and save. Then ask a question; Claude uses the connector when it helps. |
|---|---|
| Claude Code | Run the claude mcp add --transport http … command shown for the connector. |
| Other clients | Any MCP client that supports remote Streamable HTTP servers takes the address directly. A client that only runs local commands can use the mcp-remote bridge with the same address. |
| Removing one | Delete the connector in your client. Nothing is left on this side, because nothing was stored. |
Read it carefully
These connectors return published material: decisions DOHA has published, directive text and reviewed guidance. They do not predict how any case will be decided, and they are not legal advice.
What your assistant does with the results, and what it keeps of your conversation, is governed by that assistant's own terms, not by FSOtools.
Edge request counts and the rate limiter's counters are kept by the hosting platform for the length of the minute they measure. The connectors themselves keep no logs and no state, which anyone can check in their published configuration.