[rymalia]

The CLI has had AST-aware chunking for qmd query since #449 (--chunk-strategy auto), but the MCP query tool and the HTTP POST /query and /search endpoints had no way to ask for it. An agent searching an indexed code collection over MCP always got regex passage boundaries, even when the vectors had been embedded with qmd embed --chunk-strategy auto.

This PR adds an optional chunkStrategy: "auto" | "regex" to all three surfaces and forwards it to store.search(), which already supports the option.

Changes

• MCP query tool: the schema gains chunkStrategy (z.enum(["auto", "regex"]).optional()), and the handler forwards it next to filter, intent and rerank. If it's omitted, behavior is unchanged (regex).
• Server instructions: one line tells agents to pass chunkStrategy: 'auto' for indexed code.
• HTTP POST /query and /search: accept the same field. Any value other than "auto" or "regex" gets a 400 {"error": "Invalid field: chunkStrategy (must be \"auto\" or \"regex\")"}, in the same style as the existing filter validation.
• Docs: a README MCP-parameter row, a short note in skills/qmd/SKILL.md, and an [Unreleased] changelog entry.

Tests

test/mcp.test.ts adds a TypeScript fixture whose function boundaries put the query terms in different passages under each strategy. regex chunking selects the passage containing alpha, and auto (AST) selects the one containing beta/gamma. Each test checks which passage was selected, not just that the call succeeded:

• MCP query tool: regex and auto return different lines and snippets.
• REST /query and /search: the same comparison for valid values, plus a 400 with the exact error body for an invalid value.

I checked that the tests fail without the change: removing either forwarding line (MCP or REST) makes the matching test fail.

The fixture relies on the default chunk size and overlap (3,600 / 540 characters) and the 300-character snippet window, so a change to those constants would need the fixture adjusted.

Verification

• npx tsc --noEmit -p tsconfig.json: clean
• npx oxlint: 0 diagnostics
• npx vitest run test/mcp.test.ts (Node 24.22.1): 74 passed
• bun test --timeout 60000 --preload ./src/test-preload.ts test/mcp.test.ts (Bun 1.4.4): 74 passed
• Merges cleanly with the open PRs that also touch src/mcp/server.ts (fix(mcp): reject malformed REST /query searches with 400 (#998) #1002, fix(mcp): stream SSE responses from the HTTP adapter #1036, feat(mcp): opt-in compact response mode (QMD_MCP_COMPACT=1 / --compact) #1047, Export QMD MCP server factory #1057, feat(mcp): report document and embedding counts in HTTP health #1059); checked with git merge-tree.
• Independent of fix(mcp): bound HTTP request bodies and per-request collection lists #1091 (HTTP request body limits). The two merge cleanly in either order, and with both merged test/mcp.test.ts passes 78/78.
• CI: everything passes except Bun (ubuntu-latest), which fails the same two SQLite heap-limit tests as the unrelated fix(deps): bump @modelcontextprotocol/server to 2.3.1 #1085 (doctor vector sampling … 32 MiB SQLite budget and legacy fingerprint adoption … 16 MiB SQLite budget, both asserting limitBinds under Bun on Linux). main passed that job on Bun 1.4.3; both PRs ran on Bun 1.4.4, because the workflow installs bun-version: latest. Neither PR touches the store or doctor code.