NeuroDock Guardrail — Claude Code MCP server: what it does
Flags rumination loops, unregulated hyperfocus and sycophancy in AI conversations. It advises, it never silently blocks.
claude mcp add neurodock-mcp-guardrail -- uvx neurodock-mcp-guardrailWhat NeuroDock Guardrail does
NeuroDock Guardrail is a clinical MCP server for NeuroDock. It watches for three patterns that plain LLM interaction tends to amplify for neurodivergent users: OCD-style rumination loops, ADHD-style hyperfocus, and over-validation. All three detectors are live as of v0.0.2.
The server exposes three tools. check_rumination uses a word-overlap Jaccard heuristic with a 90-minute window, a count of 3 and similarity 0.55. check_hyperfocus works off an elapsed-time threshold tied to end of day, at 60, 90 and 120 minutes. check_sycophancy matches against four patterns at similarity 0.5. Those thresholds are defaults rather than prescriptions, and the heuristic sources sit in src/neurodock_mcp_guardrail/heuristics/ where you can read them.
Nothing is persisted. No SQLite, no JSONL, no cache that outlives a call — callers hand over the whole history themselves. There are no network sockets and no telemetry, and logs carry only tool-invocation metadata. You start it over stdio with the neurodock-mcp-guardrail console script, so uvx has to be on the machine. The license is AGPL-3.0-or-later, which matters if you plan to fold this into a product.
What it helps with
- spot a conversation that keeps circling the same wording
- catch a session that has run well past the end of the day
- tell real feedback apart from empty agreement
- check a chat history without storing it server-side
- see which rule fired and why
Example prompts for the agent
- Run
check_ruminationover this conversation and tell me if I'm looping - How long have I been working? Call
check_hyperfocus - Check the last few replies for sycophancy with
check_sycophancy
How to install NeuroDock Guardrail
Platform CLI
claude mcp add neurodock-mcp-guardrail -- uvx neurodock-mcp-guardrailuvx runs the command named after the package; if the README names another one, use uvx --from neurodock-mcp-guardrail <command>
Platform CLI
codex mcp add mcp-guardrail -- uvx neurodock-mcp-guardrailuvx runs the command named after the package; if the README names another one, use uvx --from neurodock-mcp-guardrail <command>
Config file
{
"mcpServers": {
"mcp-guardrail": {
"command": "uvx",
"args": [
"neurodock-mcp-guardrail"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from neurodock-mcp-guardrail <command>
Config file
{
"servers": {
"mcp-guardrail": {
"type": "stdio",
"command": "uvx",
"args": [
"neurodock-mcp-guardrail"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from neurodock-mcp-guardrail <command>
Platform CLI
gemini mcp add mcp-guardrail uvx neurodock-mcp-guardrailuvx runs the command named after the package; if the README names another one, use uvx --from neurodock-mcp-guardrail <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about NeuroDock Guardrail
How do I install NeuroDock Guardrail in Claude Code?
Run claude mcp add neurodock-mcp-guardrail -- uvx neurodock-mcp-guardrail. The server then starts over stdio through uvx.
What does NeuroDock Guardrail do?
It runs conversation history through three detectors: rumination, hyperfocus and sycophancy. Each call returns an advisory signal instead of a block.
Does the server keep my data?
No. It stores nothing between calls and writes no user content to logs, only tool-invocation metadata. The caller supplies the full history.
