veilSense / Feature Deep Dive
Tooling & MCP

Connect any tool.
Let the AI actually use it.

veilSense speaks the Model Context Protocol in both directions. Point it at any MCP server and the chat LLM calls its tools live — you decide exactly which ones. Or expose your workspace Brain as an MCP server, and plug Claude Code, Codex, or Cursor straight into your team's knowledge.

Request Demo ← Back to overview
veilsense.company.local / assistants / tool-assistant
Transport
http
Server URL
https://mcp.internal-tools.company/
Auth header
•••••••••••••• (encrypted)
Status
✓ Connected — 6 tools discovered
Allowlist
✓ search_tickets
✓ get_ticket_details
  create_ticket
  delete_ticket
✓ list_projects
Explicit allowlisting

Not "whatever the server exposes." Only what you approve.

Test Connection discovers every tool an MCP server offers. You then pick exactly which ones the LLM is allowed to call — the rest stay invisible to the model, even if the server technically supports them.

  • ✓
    Per-assistant tool allowlist, enforced server-side before any call
  • ✓
    Auth headers and stdio environment variables encrypted at rest
  • ✓
    Works over HTTP (streamable) or a local stdio command
  • ✓
    Servers that need an interactive sign-in (for example Atlassian/Jira) connect with OAuth; tokens refresh automatically
🎫
Internal Ticketing
HTTP · 2 of 6 tools allowed
Active
🗄️
Data Warehouse
stdio · 1 of 3 tools allowed
Active
📊
Analytics API
HTTP · 4 of 4 tools allowed
Active
Live in the conversation

The LLM calls tools mid-answer, not on a separate screen.

Attach a Tool Assistant to a chat and the model can call its tools as part of answering — reading a live ticket, querying a warehouse, checking an API — and use the result in the same response. A sequential-call cap keeps the loop from running away; the same mechanism powers web search with its own, higher cap.

  • ✓
    Tool calls happen inline during a normal chat turn
  • ✓
    Attach multiple Tool Assistants to run several MCP servers in one conversation
  • ✓
    Available in chat, ad-hoc runs, and as a step inside the Workflow Engine
  • ✓
    A sequential tool-call cap prevents runaway loops, tuned separately for web search
1User asks: "What's the status of ticket #4471?"
2Model calls get_ticket_details
3MCP server returns live ticket data
4Model answers using the real, current status
4 of 5 allowed sequential tool calls used
The other direction

Your Brain, as an MCP server.

veilSense exposes one workspace's Brain at a single /mcp endpoint. Add it to Claude Code, Codex, or Cursor in one line, authenticated with a Personal Access Token — the workspace and permissions are baked into the token, a sysadmin switch governs the whole feature, and requests are rate-limited.

  • ✓
    Scoped tokens: brain:read, brain:read:raw to bypass anonymization, brain:write to file entries back
  • ✓
    Every write still runs through the workspace PII review gate — nothing skips it
  • ✓
    Standard JSON-RPC over HTTP (Streamable HTTP transport) — no plugin, no bridge
$ claude mcp add veilsense --transport http \
https://veilsense.company.local/mcp \
--header "Authorization: Bearer vs_pat_••••"
✓ veilSense Brain — connected · 7 tools available
Two ways to wire in a coding agent

Grounded context — or managed dev work.

Claude Code and Codex plug into veilSense for two different jobs. Point them at the Brain for context while they code, or let veilSense hand them decomposed implementation tasks and take back the code plus the docs.

  • ✓
    Knowledge — search_brain, get_entry, get_backlinks: answers about your architecture, decisions, and conventions come from the team's real Brain, not a guess
  • ✓
    Write-back — create_entry, append_to_entry: a decision made in the coding session is filed straight into Brain
  • ✓
    Dev task board — list_dev_tasks / start_dev_task / complete_dev_task: veilSense decomposes the work, the agent picks it up and reports progress and documentation back, with the human review gate at the end
🧠
Brain context
search_brain · get_entry · get_backlinks
Available
✍️
Write-back
create_entry · append_to_entry — via PII gate
Available
📋
Dev task board
list_dev_tasks · start · complete
Preview
How it works

From an MCP server to a live tool call.

1

Point at your MCP server

HTTP URL with an auth header or an OAuth sign-in, or a local stdio command with environment variables — both encrypted at rest.

2

Test, discover, allowlist

Test Connection lists every tool the server offers. Check the ones the LLM is allowed to call.

3

Attach it anywhere

Chat, ad-hoc execution, or a node in a Workflow — the LLM calls the tool live, and uses the result in its answer.

Your systems. Your tools. Your rules.

Connect any MCP server and decide exactly what the AI is allowed to touch — or open your Brain to Claude Code and Codex with a scoped token.

Get in Touch Explore all features