Codebase MCP
An MCP server for codebase context
Schemyx uses MCP as the bridge between your local codebase context and the AI tools doing the work. Agents can resolve concepts, fetch recipes, inspect review items, and fall back to source only when needed.
The problem
Codebase context is too large to send on every request.
A mature product has too many routes, components, services, models, tokens, and standards to paste into every agent prompt. Full-context workflows become slow, noisy, and expensive.
Schemyx approach
MCP lets Schemyx serve the smallest useful context.
Agents ask for the concept they need, Schemyx resolves the key, and MCP returns a focused recipe or context pack. The source file remains available as a safe fallback.
Why teams use it
Less drift, faster onboarding, and the same context across agents.
Outcome
Segmented instead of monolithic
Schemyx writes manifest, graph, index, recipe, and review files so agents can retrieve one slice instead of a giant file.
Outcome
Stable keys for repeat work
Common requests like button styles, signup flow, admin route, or user model resolve to stable keys that can be reused.
Outcome
Built for current and future workflows
The same MCP context can support agent generation today and visual codebase maps, onboarding, and approval workflows later.
Lookup examples
The useful detail lives behind stable keys.
Schemyx is built for exact reads. Agents can ask for a concept, resolve a key, then fetch a tiny context response before they write.
.schemyx/codebase/manifest.json
Bundle manifest
Scan metadata, schema versions, extractor versions, stats, and generated paths.
.schemyx/codebase/index/terms.json
Term index
Find recipes by human search terms and agent intent.
.schemyx/codebase/recipes/style.jsonl
Style recipes
Fetch theme tokens, component style definitions, class expressions, and UI evidence.
.schemyx/codebase/review/items.json
Review items
Tell agents when a pattern needs team approval before becoming canonical.
Workflow
From codebase scan to agent-safe generation.
Schemyx keeps the heavy context work out of the prompt and turns repeated decisions into reusable lookup responses.
01
Build the bundle
Generate Schemyx codebase context locally.
02
Search through MCP
Agents call the MCP tools to resolve the right key.
03
Fetch exact context
The server returns concise JSON recipes with dependencies and provenance.
04
Read source only when needed
The fallback read is explicit, scoped, and traceable.
Who it is for
Built for teams and builders who want AI work to stay coherent.
Audience
Teams with large repos
Reduce repeated repo scans and oversized prompts.
Audience
MCP adopters
Use MCP for more than docs. Serve exact source-derived codebase context.
Audience
Agent platform builders
Give every tool a predictable retrieval layer for codebase facts.
FAQ
Questions teams ask before trusting agent context.
Why use MCP for codebase context?
MCP gives agents a consistent tool interface for search, recipe lookup, context packs, graph reads, review items, and fallback source reads.
Is the context one giant JSON file?
No. Schemyx segments context into manifest, graph, index, recipe, and review files so agents can fetch focused slices.
Can the bundle be committed to a repo?
Yes. Local bundles are designed to be portable. Teams can decide what to commit and what to regenerate.
How does this reduce drift?
Agents retrieve the same current rules for repeated tasks instead of learning conventions differently in each session.
Next step
Give every agent the same source of truth.
Request beta access or book a call to map Schemyx against your current stack, team standards, and agent workflow.