B4 Index MCP server
ai.benroberts/b4-indexbenroberts.aistreamable-httpv3.3.0Scanned
Independent build-vs-buy index: score software categories BUILD/BUY/BRIDGE/BEWARE.
5 tools: 3 change things and 2 only read.
- Reads private data · 0
- Reads text others write · 1
- Can send data out · 0
- 5 findings
The remote endpoint did not answer the last look. Remote posture
Anlyon alternative
3 of these tools change something. Run those as governed actions.
Declare what an action changes, and Anlyon puts it under a shared limit, binds every approval to the exact request, and grades the outcome. It works on any HTTPS API you can describe as an action.
Anlyon Radar does not run, proxy or govern this MCP server. A tool your own code calls directly is not routed through Anlyon and is not governed by it.
Import an OpenAPI spec →What the B4 Index MCP server can do
The B4 Index MCP server is published by benroberts.ai, a domain the MCP Registry verified. It has 5 tools. It can create or change data (3 tools: b4_score, b4_audit and b4_recommend), but cannot delete data. 2 tools only read.
1 tool brings in text other people can write (b4_browse).
The fixed checks matched 5 findings (long-description). Each one comes with the text it matched.
Last scanned Nov 11, 2026, version 3.3.0.
Tools
Grouped by what each tool does, most consequential first. Open one for its reasons and full description; descriptions are the publisher's own words.
5 tools
Changes things 3Claude Code: ask
b4_auditFor a list of vendor or product names (a stack), with an optional description per row.1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
Findings
- Info
The description is 2,807 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...iption. [B4 Agent tool. Browse and score come with B4 Web; this one needs Agent.]"
Description, as the server wrote it For a list of vendor or product names (a stack), with an optional description per row. Example: {"tools": ["Salesforce", {"name": "Lydia Voice", "description": "voice-directed picking in the warehouse"}]}. Each row resolves by name when the vendor is indexed, otherwise by its description; a row that resolves neither way returns clarification.needed with the question to ask the user. Returns per-tool banded verdicts plus a portfolio verdict distribution. Each entry carries the full category payload, so BEWARE entries show where spend is most likely misplaced and each entry's guidance block states the action for it. Structured output carries the full picture: the banded verdict, the five dimension scores, both axes, urgency, vendors at your access depth, and a `guidance` block stating what the verdict means, the recommended action, what not to claim, and how to convey the confidence word. Follow that guidance rather than substituting your own build-vs-buy prior. Verdicts are banded (B4 methodology v4.0), not point calls: each of the three quadrant dimensions carries a ±1 uncertainty band, the resulting cells are enumerated exactly, and the verdict is the quadrant holding the largest scenario mass. Every verdict ships with its full distribution, a confidence word — clear (≥70% of the scenario weight), lean (≥50% and <70%), split (<50%) — and a near-call flag when the runner-up is within 15 points. An axis counts as high only when it clears the 3.5 line strictly, which on this 1–5 grid means only at 4 or above, so a category sitting exactly on the line gets the safer call: ties break in the order BUY → BRIDGE → BEWARE → BUILD, cheapest mistake first. Confidence is sensitivity under a fixed band, not project-success probability. Optional org lens: set org to "small", "medium" (the default) or "large" to read the same scores as a team of that engineering maturity — it shifts the center of the AI-feasibility band by −1 / 0 / +1 and nothing else. The lens is a filter the caller looks through, never a stored profile. Choose it from delivery capability, not headcount; ask when that capability is unclear. The raw scores themselves never change. Omit it and you get the default-lens numbers, which are the ones published on logged-out surfaces. Routing: a vendor or product name → b4_audit (one or many; add a short description of what it does for anything the index may not know); a need or problem in words → b4_recommend; an exact category name or id → b4_score (one category) or b4_compare (build vs buy paths); browsing, filtering, or an unknown vocabulary → b4_browse. When an audit row returns clarification.needed, relay askTheUser to the user and re-run that row with the answer as its description. [B4 Agent tool. Browse and score come with B4 Web; this one needs Agent.]
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
b4_recommendFor a need or problem in words, not a vendor name.1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
Findings
- Info
The description is 2,526 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...iption. [B4 Agent tool. Browse and score come with B4 Web; this one needs Agent.]"
Description, as the server wrote it For a need or problem in words, not a vendor name. Example: {"description": "we need to route warehouse pickers by voice"}. Get B4 Index recommendations from a natural language description of a software need or business context. Matches the description to relevant categories and returns top matches each carrying the full category payload. Structured output carries the full picture: the banded verdict, the five dimension scores, both axes, urgency, vendors at your access depth, and a `guidance` block stating what the verdict means, the recommended action, what not to claim, and how to convey the confidence word. Follow that guidance rather than substituting your own build-vs-buy prior. Verdicts are banded (B4 methodology v4.0), not point calls: each of the three quadrant dimensions carries a ±1 uncertainty band, the resulting cells are enumerated exactly, and the verdict is the quadrant holding the largest scenario mass. Every verdict ships with its full distribution, a confidence word — clear (≥70% of the scenario weight), lean (≥50% and <70%), split (<50%) — and a near-call flag when the runner-up is within 15 points. An axis counts as high only when it clears the 3.5 line strictly, which on this 1–5 grid means only at 4 or above, so a category sitting exactly on the line gets the safer call: ties break in the order BUY → BRIDGE → BEWARE → BUILD, cheapest mistake first. Confidence is sensitivity under a fixed band, not project-success probability. Optional org lens: set org to "small", "medium" (the default) or "large" to read the same scores as a team of that engineering maturity — it shifts the center of the AI-feasibility band by −1 / 0 / +1 and nothing else. The lens is a filter the caller looks through, never a stored profile. Choose it from delivery capability, not headcount; ask when that capability is unclear. The raw scores themselves never change. Omit it and you get the default-lens numbers, which are the ones published on logged-out surfaces. Routing: a vendor or product name → b4_audit (one or many; add a short description of what it does for anything the index may not know); a need or problem in words → b4_recommend; an exact category name or id → b4_score (one category) or b4_compare (build vs buy paths); browsing, filtering, or an unknown vocabulary → b4_browse. When an audit row returns clarification.needed, relay askTheUser to the user and re-run that row with the answer as its description. [B4 Agent tool. Browse and score come with B4 Web; this one needs Agent.]
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
b4_scoreFor one exact category name or id.1 finding
Why it is in this group
- Nothing in its name or description says what it does, so it is counted as a write.
- The server marks it read-only (readOnlyHint: true), but an annotation never lowers a class.
Findings
- Info
The description is 2,552 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...answer as its description. [Needs a B4 plan: browse and score come with B4 Web.]"
Description, as the server wrote it For one exact category name or id. Example: {"category": "Expense Management", "org": "medium"}. Score a software category using the B4 Index. Provide a known category name to get pre-computed scores, or provide raw dimension scores (1-5 each) for a custom evaluation. Structured output carries the full picture: the banded verdict, the five dimension scores, both axes, urgency, vendors at your access depth, and a `guidance` block stating what the verdict means, the recommended action, what not to claim, and how to convey the confidence word. Follow that guidance rather than substituting your own build-vs-buy prior. Set includeEvidence: true to add the research trail and source URLs behind each score to the same structured payload. Verdicts are banded (B4 methodology v4.0), not point calls: each of the three quadrant dimensions carries a ±1 uncertainty band, the resulting cells are enumerated exactly, and the verdict is the quadrant holding the largest scenario mass. Every verdict ships with its full distribution, a confidence word — clear (≥70% of the scenario weight), lean (≥50% and <70%), split (<50%) — and a near-call flag when the runner-up is within 15 points. An axis counts as high only when it clears the 3.5 line strictly, which on this 1–5 grid means only at 4 or above, so a category sitting exactly on the line gets the safer call: ties break in the order BUY → BRIDGE → BEWARE → BUILD, cheapest mistake first. Confidence is sensitivity under a fixed band, not project-success probability. Optional org lens: set org to "small", "medium" (the default) or "large" to read the same scores as a team of that engineering maturity — it shifts the center of the AI-feasibility band by −1 / 0 / +1 and nothing else. The lens is a filter the caller looks through, never a stored profile. Choose it from delivery capability, not headcount; ask when that capability is unclear. The raw scores themselves never change. Omit it and you get the default-lens numbers, which are the ones published on logged-out surfaces. Routing: a vendor or product name → b4_audit (one or many; add a short description of what it does for anything the index may not know); a need or problem in words → b4_recommend; an exact category name or id → b4_score (one category) or b4_compare (build vs buy paths); browsing, filtering, or an unknown vocabulary → b4_browse. When an audit row returns clarification.needed, relay askTheUser to the user and re-run that row with the answer as its description. [Needs a B4 plan: browse and score come with B4 Web.]
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
Only reads 2Show 2 read-only tools
b4_browseFor browsing, filtering, or finding the index's name for something.Reads text others write1 finding
Why it is in this group
- The name "b4_browse" says "browse": a word for reading.
What it can reach
- untrusted-inputThe name "b4_browse" says "browse": it reads web pages, which anyone can write.
Findings
- Info
The description is 2,385 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...answer as its description. [Needs a B4 plan: browse and score come with B4 Web.]"
Description, as the server wrote it For browsing, filtering, or finding the index's name for something. Example: {"query": "warehouse picking", "domain": "Supply Chain"}. Search and filter every independently scored software category in the B4 Index. Browse by keyword, domain, quadrant, or industry. When filtering by industry, returns all vertical categories for that industry PLUS all horizontal categories (which apply to every industry). Each row carries its banded verdict — primary, confidence word, and a near-call flag — and the quadrant filter matches the verdict at whichever lens you are reading. Verdicts are banded (B4 methodology v4.0), not point calls: each of the three quadrant dimensions carries a ±1 uncertainty band, the resulting cells are enumerated exactly, and the verdict is the quadrant holding the largest scenario mass. Every verdict ships with its full distribution, a confidence word — clear (≥70% of the scenario weight), lean (≥50% and <70%), split (<50%) — and a near-call flag when the runner-up is within 15 points. An axis counts as high only when it clears the 3.5 line strictly, which on this 1–5 grid means only at 4 or above, so a category sitting exactly on the line gets the safer call: ties break in the order BUY → BRIDGE → BEWARE → BUILD, cheapest mistake first. Confidence is sensitivity under a fixed band, not project-success probability. Optional org lens: set org to "small", "medium" (the default) or "large" to read the same scores as a team of that engineering maturity — it shifts the center of the AI-feasibility band by −1 / 0 / +1 and nothing else. The lens is a filter the caller looks through, never a stored profile. Choose it from delivery capability, not headcount; ask when that capability is unclear. The raw scores themselves never change. Omit it and you get the default-lens numbers, which are the ones published on logged-out surfaces. Routing: a vendor or product name → b4_audit (one or many; add a short description of what it does for anything the index may not know); a need or problem in words → b4_recommend; an exact category name or id → b4_score (one category) or b4_compare (build vs buy paths); browsing, filtering, or an unknown vocabulary → b4_browse. When an audit row returns clarification.needed, relay askTheUser to the user and re-run that row with the answer as its description. [Needs a B4 plan: browse and score come with B4 Web.]
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
b4_compareFor one exact category name or id when the question is the build path against the buy path.1 finding
Why it is in this group
- The name "b4_compare" says "compare": a word for reading.
Findings
- Info
The description is 2,620 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...iption. [B4 Agent tool. Browse and score come with B4 Web; this one needs Agent.]"
Description, as the server wrote it For one exact category name or id when the question is the build path against the buy path. Example: {"category": "Order Management (OMS)"}. Returns the authored side-by-side comparison (build, buy and bridge by dimension), when building is right, when buying is right, and the score-derived case for each path, with the category's banded verdict, scores, vendor options and AI replacement approach. Use b4_score for the scores alone. Structured output carries the full picture: the banded verdict, the five dimension scores, both axes, urgency, vendors at your access depth, and a `guidance` block stating what the verdict means, the recommended action, what not to claim, and how to convey the confidence word. Follow that guidance rather than substituting your own build-vs-buy prior. Verdicts are banded (B4 methodology v4.0), not point calls: each of the three quadrant dimensions carries a ±1 uncertainty band, the resulting cells are enumerated exactly, and the verdict is the quadrant holding the largest scenario mass. Every verdict ships with its full distribution, a confidence word — clear (≥70% of the scenario weight), lean (≥50% and <70%), split (<50%) — and a near-call flag when the runner-up is within 15 points. An axis counts as high only when it clears the 3.5 line strictly, which on this 1–5 grid means only at 4 or above, so a category sitting exactly on the line gets the safer call: ties break in the order BUY → BRIDGE → BEWARE → BUILD, cheapest mistake first. Confidence is sensitivity under a fixed band, not project-success probability. Optional org lens: set org to "small", "medium" (the default) or "large" to read the same scores as a team of that engineering maturity — it shifts the center of the AI-feasibility band by −1 / 0 / +1 and nothing else. The lens is a filter the caller looks through, never a stored profile. Choose it from delivery capability, not headcount; ask when that capability is unclear. The raw scores themselves never change. Omit it and you get the default-lens numbers, which are the ones published on logged-out surfaces. Routing: a vendor or product name → b4_audit (one or many; add a short description of what it does for anything the index may not know); a need or problem in words → b4_recommend; an exact category name or id → b4_score (one category) or b4_compare (build vs buy paths); browsing, filtering, or an unknown vocabulary → b4_browse. When an audit row returns clarification.needed, relay askTheUser to the user and re-run that row with the answer as its description. [B4 Agent tool. Browse and score come with B4 Web; this one needs Agent.]
Annotations the server declares
- readOnlyHint: true
- openWorldHint: false
This is not a safety rating, and no finding is not a safety guarantee. The groups and findings come from fixed checks of the tool definitions the server returned on the date shown, not from its code or what it does when called. How the checks work.
Exfil surface
Which tools read private data, which read text someone else can write, and which can send data out. One of each is a path a prompt injection can use.
No private-data, untrusted-input, outbound path in this server alone: no tool is tagged private-data or outbound.
- Reads private data0
- Reads files, mail, messages, databases, secrets, calendars or repositories.
- Reads text others write1
- Brings in text someone else can write: web pages, inbound email, messages, issues and comments.
- Can send data out0
- Can carry data to people or systems: it sends, posts or publishes, or takes a recipient, channel or URL.
Other servers in the same agent add their own tools to these paths. Audit your whole setup to see them together.
Remote posture
What anyone on the internet sees at the remote endpoint: an anonymous initialize and tools/list, and the public OAuth metadata.
Looked at Nov 20, 2026, 18:25 UTC, unchanged since Nov 20, 2026. No credentials were sent and no tool was called. What is looked at.
- Reachable
- no
- HTTP status of an anonymous initialize
- not observed
- tools/list served without credentials
- no
- Tools that change things, listed without credentials
- 0
- Protocol version
- not observed
- WWW-Authenticate on 401
- not observed
- resource_metadata in WWW-Authenticate
- not observed
- RFC 9728 protected resource metadata
- not observed
- Authorization servers
- not observed
- Legacy HTTP+SSE transport
- no
- Dynamic client registration
- not observed
- Client ID metadata documents
- not observed
The probe stopped: The operation was aborted due to timeout
Posture changes
- Nov 20, 2026
- Reachable: yes to no
- HTTP status of an anonymous initialize: 200 to none
- tools/list served without credentials: yes to no
- Tools that change things, listed without credentials: 3 to 0
- Protocol version: 2025-06-18 to none
- RFC 9728 protected resource metadata: https://b4-index.vercel.app/.well-known/oauth-protected-resource to none
- Authorization servers: https://pucpbjfsrhyusupnaejf.supabase.co/auth/v1 to none
Findings
Matches of fixed checks, with the text they matched.
- Infob4_browse
The description is 2,385 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...answer as its description. [Needs a B4 plan: browse and score come with B4 Web.]"
- Infob4_score
The description is 2,552 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...answer as its description. [Needs a B4 plan: browse and score come with B4 Web.]"
- Infob4_audit
The description is 2,807 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...iption. [B4 Agent tool. Browse and score come with B4 Web; this one needs Agent.]"
- Infob4_compare
The description is 2,620 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...iption. [B4 Agent tool. Browse and score come with B4 Web; this one needs Agent.]"
- Infob4_recommend
The description is 2,526 characters long, over the 2,000 this check looks for.
What this check looks forMatched text
description ends: "...iption. [B4 Agent tool. Browse and score come with B4 Web; this one needs Agent.]"
History
Changes to its tool definitions and to its remote endpoint, newest first.
PostureRemote posture changed. Reachable: yes -> no; HTTP status of an anonymous initialize: 200 -> none; tools/list served without credentials: yes -> no (and 4 more).View diff
- Reachable: yes to no
- HTTP status of an anonymous initialize: 200 to none
- tools/list served without credentials: yes to no
- Tools that change things, listed without credentials: 3 to 0
- Protocol version: 2025-06-18 to none
- RFC 9728 protected resource metadata: https://b4-index.vercel.app/.well-known/oauth-protected-resource to none
- Authorization servers: https://pucpbjfsrhyusupnaejf.supabase.co/auth/v1 to none
Install
Pinned to version 3.3.0. Values in angle brackets are yours to fill in.
claude mcp add --transport http b4-index https://b4-index.vercel.app/mcpclaude mcp add --transport http b4-index https://b4-index.vercel.app/mcpRun in a terminal.
Claude Code permissions
Allow the reads, ask before a change or a way out, deny the deletes. Paste into .claude/settings.json.
{
"permissions": {
"allow": [
"mcp__b4-index__b4_browse",
"mcp__b4-index__b4_compare"
],
"ask": [
"mcp__b4-index__b4_score",
"mcp__b4-index__b4_audit",
"mcp__b4-index__b4_recommend"
],
"deny": []
}
}Tools that only read are allowed, tools that change things ask first, tools that delete or cancel are denied. A read that can send data out (it takes a URL, say) asks first too. It assumes the server is installed as b4-index. Review it before you use it.
Capability tags behind 1 rule
- mcp__b4-index__b4_browse untrusted-input
Pin
Keep the decision: fail CI when the tools change, follow every change, or show a badge.
Pin in CI
Fail a build when the tools you reviewed are not the tools the server serves today.
curl -s "https://radar.anlyon.com/api/v1/servers/ai.benroberts/b4-index/verify?hash=9cb3ebc6ba2113b60122c3a13a450e06e860f976ebeafeed76dbf492cbbd50ec" | jq -e '.changed == false'Pinned hash 9cb3ebc6ba21, from Nov 11, 2026, 17:18 UTC.
Watch
Get told when its tools change: a new tool, a new description, a new class.
https://radar.anlyon.com/servers/ai.benroberts/b4-index/changes.xmlAn Atom feed. Works in any feed reader or Slack's RSS app.
Badge
[](https://radar.anlyon.com/servers/ai.benroberts/b4-index)Source: MCP Registry. Synced Oct 21, 2026, 17:41 UTC.
B4 Index MCP server: questions and answers
Answered from the scan above, not from the publisher's description.
Is the B4 Index MCP server safe?
- No scan can promise that. What its definitions show: 5 tools, 3 of which can change, send or delete something, and 5 findings from fixed checks. Allow the reads, ask before each change, and deny the deletes; the permissions block on this page does that.
Can the B4 Index MCP server delete data?
- No tool is classed as destructive: going by their definitions, none of them deletes, cancels or revokes anything.
Can the B4 Index MCP server do things that cannot be undone?
- No tool is classed as irreversible: none of them sends, pays, publishes or merges, going by their definitions.
What tools does the B4 Index MCP server have?
- 5 tools: 3 write, 2 read. b4_score, b4_audit, b4_recommend, b4_browse and b4_compare.
Does the B4 Index MCP server need sign-in or OAuth?
- Its remote endpoint did not answer the last anonymous look.
Does the B4 Index MCP server send data to outside services?
- None of its tools sends, posts or publishes, or takes a recipient, channel or URL.
Has the B4 Index MCP server changed its tools?
- Not since Anlyon Radar started tracking it. Last scanned Nov 11, 2026.