Docs search tools

search_docs, ask_docs, and read_doc MCP tools plus memtrace://docs/* resources — query the Memtrace documentation from your coding agent without leaving the conversation.

Overview#

Memtrace exposes its documentation to coding agents through three MCP tools and a set of memtrace://docs/* resources. The tools call the same hosted backend that powers the Ask AI widget on memtrace.io/docs — semantic search over the full docs corpus, grounded answers with citations, and strict guardrails that keep responses on-topic.

Core code-graph tools remain fully offline. Docs tools degrade gracefully when the hosted API is unreachable — they return a clear error instead of blocking the rest of the MCP server.

Four agent skills route documentation questions to these tools: memtrace-docs (umbrella), memtrace-docs-ask, memtrace-docs-search, and memtrace-docs-read — installed with Agent skills.

Tools#

Tool	Purpose
`search_docs`	Semantic search — returns ranked doc chunks with slug, title, H2, and excerpt
`ask_docs`	RAG Q&A — returns {answer, citations[], refused} grounded in docs only
`read_doc`	Full plain-text content of a docs page by slug

Example: search_docs

# In your agent conversation:
search_docs(query="how do I set up fleet coordination", limit=5)

# Returns ranked chunks like:
# { ok: true, results: [{ slug: "features/fleet", pageTitle: "...", excerpt: "..." }] }

Example: ask_docs

ask_docs(question="What ports does memtrace start use?")

# Returns:
# { ok: true, answer: "...", citations: ["cli/start"], refused: false }

Example: read_doc

read_doc(slug="getting-started")

# Returns:
# { ok: true, slug: "getting-started", title: "Getting started", body: "..." }

MCP resources#

The MCP server advertises every public docs page as a resource at memtrace://docs/<slug>. Agents that support MCP resources can browse the catalog via list_resources and read any page with read_resource — useful when you want the full page text without a tool round-trip.

RESOURCE URIS

Examples: memtrace://docs/getting-started, memtrace://docs/cli/start, memtrace://docs/mcp/overview.

Web Ask AI widget#

On memtrace.io/docs, a floating Ask AI button opens a chat panel powered by the same RAG backend. Answers stream in with markdown rendering (code blocks, diagrams) and inline citations back to source pages. The ⌘K search overlay also includes semantic results and an Ask AI about this query shortcut.

No Memtrace install is required to use the web widget — it is available to every docs visitor.

Configuration#

Docs tools call the hosted API at https://memtrace.io by default. Override with:

environment

MEMTRACE_DOCS_API_URL=https://memtrace.io

The hosted API endpoints are:

POST /api/docs-search — semantic search
POST /api/docs-chat — RAG chat (streaming for web, stream: false for MCP)
GET /api/docs/index — list all doc pages (for resource enumeration)
GET /api/docs/<slug> — full page text

See also MCP server overview for transport and client setup.