NeuroDock Cognitive Graph — Claude Code MCP Server Guide
Local typed-edge graph that holds persistent memory about projects, team members, and decisions. Runs inside SQLite with zero external network traffic.
claude mcp add neurodock-mcp-cognitive-graph -- uvx neurodock-mcp-cognitive-graphWhat NeuroDock Cognitive Graph does
The server provides persistent memory for coding assistants, preventing lost context during task switches. It tracks entities and relationships locally in ~/.neurodock/cognitive-graph.sqlite. You can point it elsewhere by exporting the NEURODOCK_GRAPH_DB_PATH environment variable.
Release v0.0.3 exposes four tools: recall_entity, record_fact, recall_decisions, and weekly_rollup. Relationship edges must strictly follow a closed set of eight predicates, such as depends_on, blocked_by, and decided_in. Passing an unrecognised predicate immediately fails with PREDICATE_NOT_IN_VOCABULARY.
Embeddings and vector recall are deferred to subsequent releases. The process operates entirely offline, avoids telemetry, and never makes HTTP requests to fetch referenced sources.
What it helps with
- persist project context and entity relations across work sessions
- log architectural decisions tied to a specific initiative
- inspect immediate dependencies and blockers for an active component
- assemble a seven-day activity rollup without making extra model calls
Example prompts for the agent
- Remember that Auth depends on Redis and belongs to Core
- Recall all recorded decisions for the Gateway project
- Generate a weekly rollup report for the trailing seven days
How to install NeuroDock Cognitive Graph
Platform CLI
claude mcp add neurodock-mcp-cognitive-graph -- uvx neurodock-mcp-cognitive-graphuvx runs the command named after the package; if the README names another one, use uvx --from neurodock-mcp-cognitive-graph <command>
Platform CLI
codex mcp add mcp-cognitive-graph -- uvx neurodock-mcp-cognitive-graphuvx runs the command named after the package; if the README names another one, use uvx --from neurodock-mcp-cognitive-graph <command>
Config file
{
"mcpServers": {
"mcp-cognitive-graph": {
"command": "uvx",
"args": [
"neurodock-mcp-cognitive-graph"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from neurodock-mcp-cognitive-graph <command>
Config file
{
"servers": {
"mcp-cognitive-graph": {
"type": "stdio",
"command": "uvx",
"args": [
"neurodock-mcp-cognitive-graph"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from neurodock-mcp-cognitive-graph <command>
Platform CLI
gemini mcp add mcp-cognitive-graph uvx neurodock-mcp-cognitive-graphuvx runs the command named after the package; if the README names another one, use uvx --from neurodock-mcp-cognitive-graph <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 Cognitive Graph
How to install NeuroDock Cognitive Graph in Claude Code?
Run claude mcp add neurodock-mcp-cognitive-graph -- uvx neurodock-mcp-cognitive-graph. It registers the package via uvx without manual setup.
Where does the graph store its database?
Data persists in ~/.neurodock/cognitive-graph.sqlite. Override the location with the NEURODOCK_GRAPH_DB_PATH environment variable.
Which predicates can be used for recording facts?
Version v0.0.3 restricts facts to eight predicates: mentioned_in, decided_in, reports_to, depends_on, resolved_by, blocked_by, tagged, and belongs_to.
