bootspring_brain
MCP tool for natural language routing — automatically dispatch queries to the right Bootspring capability
The bootspring_brain MCP tool gives AI coding assistants direct access to Bootspring's NLU routing engine. Instead of requiring users to know exact command names, the brain interprets natural language and dispatches to the best matching skill, workflow, agent, or pipeline target.
Overview
When integrated with an AI assistant (Claude Code, Cursor, Codex, etc.), the brain tool enables natural conversations that automatically invoke the right Bootspring capability.
User: "I want to check if this project is ready to ship"
AI Assistant invokes: bootspring_brain
→ query: "check if this project is ready to ship"
Brain returns:
→ target: deploy-readiness (skill)
→ confidence: 0.89
→ command: bootspring deploy check
AI Assistant runs the command and reports results.
Tool Schema
Input Parameters
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
query | string | Yes | — | Natural language query (1-500 characters) |
autoDispatch | boolean | No | false | If true, invoke the matched target when confidence exceeds the auto-dispatch threshold |
context | string | No | — | Optional context to improve routing accuracy (up to 2000 characters) |
Output Format
Successful Route
Low Confidence (Below minConfidence)
Auto-Dispatch Result
When autoDispatch is true and confidence exceeds the threshold (default 0.85):
When confidence is below the threshold, dispatched is false and the result includes the match but does not invoke it:
Integration with AI Assistants
Claude Code
The brain tool is registered automatically when Bootspring MCP is configured. Claude Code can invoke it in two ways:
Implicit — Claude recognizes when a user request maps to a Bootspring capability and invokes the brain behind the scenes:
User: "Can you check if there are any anti-patterns in my session?"
Claude invokes bootspring_brain:
query: "check for anti-patterns in my session"
→ target: observer-patterns
→ confidence: 0.93
→ command: bootspring observer patterns
Claude then runs the command and interprets the results.
Explicit — The user can ask Claude to use the brain directly:
User: "Use bootspring_brain to figure out what command handles cost optimization"
Claude invokes bootspring_brain:
query: "cost optimization"
→ target: cost-optimizer (skill)
→ confidence: 0.90
Cursor / Codex / Other MCP Clients
Any MCP-compatible client can invoke bootspring_brain with the same schema. The tool is listed in the MCP tool manifest alongside all other Bootspring tools.
Auto-Dispatch Behavior
Auto-dispatch controls whether the brain invokes the matched target or only returns the routing result.
| autoDispatch | Confidence | Behavior |
|---|---|---|
false | Any | Return match only, never invoke |
true | >= 0.85 | Invoke target, return match + result |
true | < 0.85 | Return match only with confidence warning |
The auto-dispatch threshold is configurable server-side (default: 0.85). When a target is auto-dispatched, the dispatched field is true and the result field contains the output of the invoked command.
Example Conversations
Build Loop
User: "What's my build progress?"
bootspring_brain → query: "What's my build progress?"
→ target: build-loop (skill), confidence: 0.94
AI runs: bootspring build status
AI responds: "You've completed 66 out of 114 tasks (58%). Currently working on bs-1078."
Code Review
User: "Review the auth module for issues"
bootspring_brain → query: "Review the auth module for issues"
→ target: code-review (workflow), confidence: 0.87
AI runs: bootspring workflow run code-review
AI responds with the review results.
Ambiguous Query
User: "Help me with the project"
bootspring_brain → query: "Help me with the project"
→ target: null, confidence below threshold
AI responds: "I wasn't sure what you'd like to do. Here are some options:
- 'Check build progress' → build status
- 'Review code quality' → quality gates
- 'View session analytics' → observer dashboard
What would you like?"
Multi-Step with Context
User: "I just finished the auth feature, what should I do next?"
bootspring_brain → query: "finished auth feature, what next"
context: "Current task: bs-1042, status: in_progress"
→ target: build-loop (skill), confidence: 0.91
AI runs: bootspring build done
AI runs: bootspring build next
AI responds: "Marked bs-1042 as complete. Your next task is bs-1043: Implement rate limiting middleware."
Error Handling
| Error Code | Cause | Response |
|---|---|---|
BRAIN_NOT_READY | Brain still loading targets on startup | Retry after 1-2 seconds |
QUERY_TOO_LONG | Query exceeds 500 characters | Shorten the query |
DISPATCH_FAILED | Auto-dispatch target returned an error | Match is returned with the dispatch error |
NO_TARGETS | No skills, workflows, or agents are configured | Run bootspring setup |