Product 14
The Collective Brain MCP Server
Make your company's knowledge something every AI tool you run can query, instead of a folder someone has to remember to open.
For technical founders and teams running an agent stack on top of a Markdown knowledge base.
The problem
Most team knowledge bases are passive. They sit in a wiki, a Confluence space or a folder of Markdown files: readable by a human who thinks to look, invisible to every AI tool that was not handed the right file at the start of the session.
The result is agents reasoning from generic assumptions because the company's specific intelligence was never made reachable, context re-established manually at the start of every session, and knowledge that compounds in documents nobody queries.
The team that delivers it
What it does
Your operational knowledge base, the SOPs, the strategic principles, the agent configurations, the financial context, is wrapped in a Model Context Protocol server and exposed over your own private network.
Any MCP-compliant client, Claude Desktop, Cursor, or a custom agent runtime, queries it live through five tools: read a file, list files, search the brain, propose an update, and get project status.
Six specialist agents keep it trustworthy. The Knowledge Base Curator prunes deprecated context and resolves contradictions. The Brain Pulse Monitor watches uptime and query latency. The Brain Git Steward version-controls every automated change. The PENDING-UPDATES Triager intercepts every proposed update for human review before it is committed. The Agent Registry Steward maintains the agent catalogue and tool permissions. The Daily Reports Compiler delivers the morning brief.
Built and proven as the operational backbone of Digital Professionals Academy: thirty-plus structured files across thirteen functional layers, six maintenance agents, a nightly self-audit and a human review gate on every proposed change. Every agent behind this site reads from it before acting.
The strategy it enables
Your AI tools reason from your specific context instead of generic assumptions, and new people onboard in hours because the company's intelligence is structured and queryable rather than tribal. Every session can also propose updates back, which a human approves and a nightly agent executes, so the knowledge base improves as a by-product of using it.
Who it is for
- Technical founders and AI engineersRunning complex agent stacks that need one company memory layer rather than pasted context.
- Remote and async engineering teamsStandardising on MCP-enabled tooling and needing internal documentation reachable from any IDE.
- Solo operators with large Markdown repositoriesWho want their tools reasoning from their own second brain instead of around it.
Pricing
Setup and licence
CHF 3,500 one-time
- Full deployment against your repository
- Private network configuration
- Five MCP tools configured
- Three months of update support
Maintenance retainer
CHF 490 / month
- Monthly health audit
- Schema updates as the protocol evolves
- Agent registry tuning
Managed infrastructure
CHF 1,990 / month
- Deployment and maintenance handled for you
- Daily monitoring and remote health checks
- Custom triage heuristics
The maintenance retainer requires an active licence. Deployment options are a private mesh network endpoint or a filesystem-mounted client, chosen against your security posture during setup.
This is the infrastructure layer underneath the rest of this catalogue, built for the Nigel Casey operating company and deployed against your own repository, your own network and your own review policy. Nothing about it is specific to language education.
Technical documentarchitecture, integrations, setup
Protocol
- Model Context Protocol, the Anthropic open standard
- Transport: HTTP, streamable
- Authentication: bearer token held in an environment variable
- Network: private mesh with valid TLS and no open firewall ports, or a filesystem-mounted client where the runtime supports it
The five exposed tools
- read_file(filename): returns the full content of one file
- list_files(): returns every Markdown filename in the knowledge base root
- search_brain(query): case-insensitive keyword search returning 200-character context excerpts
- propose_update(type, file, section, old, new, reason, source): appends a structured entry to the pending-updates queue for human review
- get_project_status(project_name): fuzzy match against the ecosystem map
Compatible clients
- Claude Desktop, Cursor, Zed
- Custom Node.js or Python agent runtimes
- Any client implementing MCP 2024-11-05 or later
Knowledge base requirements
- Markdown files in a Git repository
- Recommended: a scope contract per file and a typed update system covering new, replace, refine and reframe
- Minimum viable: 5 to 10 structured files. Optimal: 20 to 30 across functional domains
Infrastructure and security
- Host: a Mac, a Linux server or a VPS with persistent uptime. Node.js 20 or later
- Zero-trust: no public firewall ports, bearer token on every route, network ACL restricted to authorised nodes
- Every automated change is a Git commit with a structured diff and a remote backup
- Human review required before any proposed update is committed
Setup timeline
- Week 1: knowledge base audit, file structure design, scope contract definition
- Week 2: server deployment, network configuration, authentication, tool verification
- Week 3: maintenance agent configuration, nightly loop, client integration testing