Docs MCP Manifest
Configuration file for Docs MCP
| Type | object |
|---|---|
| File match |
.docs-mcp.json
|
| Schema URL | https://catalog.lintel.tools/schemas/schemastore/docs-mcp-manifest/latest.json |
| Source | https://raw.githubusercontent.com/speakeasy-api/docs-mcp/refs/heads/main/schemas/docs-mcp.schema.json |
Validate with Lintel
npx @lintel/lintel check
Docs MCP configuration file (.docs-mcp.json) that controls how documentation is chunked, tagged, and indexed for search.
Properties
Schema version. Must be '1'.
JSON Schema URI for editor validation. Typically set automatically by SchemaStore; can be added explicitly.
Default chunking strategy applied to all files in this directory tree unless overridden by a more specific rule.
3 nested properties
The heading level at which to split markdown into chunks. 'h1' splits at top-level headings, 'h2'/'h3' at progressively finer granularity, and 'file' treats the entire file as one chunk.
Maximum chunk size in characters (default: 20000). Oversized chunks are first split recursively at finer heading levels (e.g. h2→h3→h4→…→h6), preserving semantic structure and breadcrumbs. Only when no further sub-headings exist does it fall back to AST node boundary splitting.
Minimum chunk size in characters. Trailing chunks smaller than this are merged into the preceding chunk to avoid fragments.
Key-value pairs attached to every chunk produced from this directory tree. Each key becomes a filterable taxonomy dimension exposed as an enum parameter on the search tool.
Per-field configuration for taxonomy dimensions. Controls search-time behavior such as cross-language result collapsing.
Custom MCP server instructions sent to clients during initialization. Helps coding agents understand what this server provides and how to use it effectively.
Per-file-pattern overrides for chunking strategy and metadata. Evaluated top-to-bottom; last match wins.