Query Markup Documents - On-device hybrid search for markdown files with BM25, vector search, and LLM reranking
M1K3 is installed from its publisher's own source and answers where it runs, so this marketplace is not in the path of a single call. There is no address here to send one to, and a panel that pretended otherwise would be showing you an answer we made up. Install it and call it from your own client — the Installation tab has the entry for each one.
What it does
M1K3 is a local, offline AI assistant for Apple Silicon Macs. While the app is open it serves MCP over HTTP on loopback (http://127.0.0.1:4242/mcp): 18 tools, nothing hosted, and nothing leaves the machine unless you switch a network feature on.
Knowledge: search_knowledge, list_documents, get_document. Hybrid search over the notes, PDFs and call transcripts the user has indexed.
Memory: remember, recall_memory, related_memory, forget_memory, memory_stats. A temporal memory graph the user can inspect, correct and delete from.
A local brain: ask_m1k3, get_answer, list_jobs. Ask the resident model (Apple Foundation Models, Qwen 3 4B or Gemma 4 12B via MLX) and fetch long answers by job id.
Voice: speak, stop_speaking, listen. The user's voice channel, with an animated avatar, for the moments that matter. Also: list_todos, propose_todo, open_link, get_status.
Setup: in the app open Settings, then MCP Server, switch it on and copy the token; then run m1k3 login && m1k3 connect claude. Every request carries a bearer token the app mints and keeps in the Keychain. A knowledge-only stdio binary (3 tools) exists for clients that cannot speak HTTP. Full guide for every client: https://m1k3.app/agents
Free. Source-available (FSL-1.1-ALv2). No account, no telemetry. Mac App Store, Homebrew cask or DMG.
Quickstart
Install it the way its author documents:
The full guide: https://m1k3.app/agents
Where are you running it?
Every route below installs the same thing and ends at the same approval screen. This one runs on your machine: your client starts M1K3 as a process under your own user, with your files and your network, so the tool surface below is what it can do to your computer rather than to a server somewhere else. Its scan result is on the Security tab — read the surface before you approve it.
Install it the way its author documents:
The full guide: https://m1k3.app/agents
Tool surface
What the model actually sees. Descriptions are diffed on every release — see version history.
Search M1K3's stored knowledge (documents, calls, notes; hybrid retrieval when available). Returns ranked matching chunks.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| query | string | yes | the text to search for |
| limit | integer | no | max results (default 5) |
List the items M1K3 has indexed, with their ids, kinds, and titles.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| limit | integer | no | max items (default 100) |
Fetch the text of one indexed item by its id (from list_documents). Returns up to ~6000 characters per call; for longer items, a footer gives the offset to resume from — call again with that offset to page through.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| id | string | yes | the document id (UUID) |
| offset | integer | no | character offset to start from (default 0; see the resume footer) |
Ask M1K3's local brain a question. The answer is grounded in M1K3's private knowledge store with section-level citations, and may use web search if the user has it enabled in M1K3's settings. Fully local inference. Fast answers return inline; a long turn (web search on a larger brain) instead returns a job id — redeem it in a few seconds via get_answer, or by calling this tool again with just the
| Parameter | Type | Required | Notes |
|---|---|---|---|
| question | string | no | the question to ask (omit when redeeming a job_id) |
| job_id | string | no | a job id a previous ask_m1k3 call returned — fetches that result instead of asking a new question |
Fetch the result of an ask_m1k3 call that returned a job id because it was taking a while. Returns the grounded answer once ready, or asks you to poll again if M1K3 is still working. Only needed when ask_m1k3 handed back a job id.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| job_id | string | yes | the job id ask_m1k3 returned |
List recent ask_m1k3 jobs — id, state (running/done/error), and age in seconds. Use this to recover a job id you lost; redeem a done job via get_answer. Never contains answer text.
No parameter schema on file.
Store text in M1K3's memory — it becomes part of what M1K3 knows, searchable in every future conversation (the same store search_knowledge reads). Use it for durable facts, notes, summaries, decisions. Survives every session.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| title | string | yes | a short title for the entry |
| text | string | yes | the content to index |
| kind | string | no | what kind of fact this is: profile (a stable fact about the user), preference (a standing preference), decision (a decision and why), episode (something that happened), or note (anything else — the de |
Recall atomic facts M1K3 remembers about the user — the temporal memory GRAPH, separate from the document corpus search_knowledge reads. Returns the matching facts (with a similarity hint) or says so honestly when nothing is recalled. Use it to check what M1K3 has stored about a person, preference, or decision.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| query | string | yes | what to recall (a name, topic, or question) |
Recall the single best matching fact for the query, then walk M1K3's memory GRAPH one step out to its neighbours (linked or superseded facts). Use it to see how a memory connects to others — not just flat recall. Says so when there is no matching fact to anchor from.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| query | string | yes | what to anchor the graph walk on |
How many atomic facts M1K3 currently remembers (the live, non-superseded count). Handy for watching the memory graph grow as facts are remembered.
No parameter schema on file.
Permanently forget a fact M1K3 remembers — the consent primitive, the counterpart to remember. Finds the single best-matching memory for the query and HARD-DELETES it from BOTH the memory graph AND the document corpus (no residue). Irreversible: when no memory is a confident match it deletes NOTHING and tells you the closest, so a vague query can't erase the wrong fact. Repeat the remembered FACT'
| Parameter | Type | Required | Notes |
|---|---|---|---|
| query | string | yes | the fact to forget — repeat the remembered fact's text word-for-word (not your search query) for precision |
Speak text aloud through M1K3's voice (and animate the avatar). Optional emotion, which sets the face and the voice's pace (and pitch where the voice allows): happy, sad, angry, surprised, love, thinking, excited, sleepy, neutral. Pass wait:true to return once playback finishes, or after ~25s with a still-speaking note for a long read (poll get_status); default returns as speech starts. If another
| Parameter | Type | Required | Notes |
|---|---|---|---|
| text | string | yes | what to say |
| emotion | string | no | face and voice mood while speaking (optional) |
| wait | boolean | no | return after playback completes, bounded at ~25s (default false) |
Stop any in-progress speech immediately.
No parameter schema on file.
M1K3's overall status: active brain tier, TTS provider, voice tier, and the busy flags — whether M1K3 is speaking, in a conversation, using its mic, or already answering an ask_m1k3 call. Poll this before speak/listen/ask_m1k3.
No parameter schema on file.
Listen on M1K3's microphone and return the transcript once the speaker pauses (or the timeout passes). Waits for any in-progress speech to finish before opening the mic — that wait comes OUT of timeout_s (at least 5s of real listening is always kept), so the call is bounded by roughly timeout_s overall. This is how you hear the user.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| timeout_s | number | no | max seconds to listen (default 30, clamped 5-120) |
The user's todo list as M1K3 holds it. open (default) is what is on the list; pending is what M1K3 or a visiting agent has proposed and the user has not yet accepted; all includes done and dismissed. Read-only.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| state | string | no | which items to list (default open) |
PROPOSE a todo for the user. It lands in their inbox as pending, stamped with your client name; only the user can accept, complete, or dismiss it. This never opens a todo, never marks anything done, and is refused when the user has turned suggestions off or the inbox is full. Use it for something the user said they still need to do — not for your own work.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| title | string | yes | the task, in the user's terms, one line |
| note | string | no | optional context — why, or where it came from |
| due | string | no | optional ISO-8601 date (2026-09-14) or date-time |
Open a web link in M1K3's review panel on the user's screen, beside the conversation, so they can see the page. Local-first — the page opens in M1K3, nothing is sent anywhere. Argument: an http(s) URL.
| Parameter | Type | Required | Notes |
|---|---|---|---|
| url | string | yes | the http(s) URL to open |
- Every tool, no call limit
- No card, and mcprush charges nothing for it
- Source is public; its licence metadata reads NOASSERTION rather than naming a licence
- Runs on your machine — nothing of it reaches our gateway
- Nothing to cap, because mcprush bills nothing
What counts against your monthly calls
| Tool | Unit | Calls used | Out of the allowance |
|---|
Nothing here is billable: mcprush charges nothing to install M1K3 or to call it, at any volume.
Two independent axes, because powerful and malicious are different questions. The grade is threat only. The capability level is blast radius, and it is never a penalty on the grade — it is priced as one subtract-only term in the score, where you can see it.
| Term | Level | What it prices | Points |
|---|---|---|---|
| capability-exposure | high | capability blast radius (high) — client exposure if the model is manipulated | −6 |
| verification-discount | repo | publisher verification (public source) — no provenance, but the source is public and inspectable | −1 |
| coverage-honesty | source | inspection depth (source) — how much of the target the scan could see | −0 |
What the scan could actually read
A grade is only as meaningful as its coverage, so the scanner publishes its own depth before it publishes its result.
No tools were enumerated, so prompt-injection, capability and toxic-flow analysis had no tool surface to inspect. To grade a package’s real runtime tools, scan the running server: --command "npx -y <package>".
Capability — what it could do if the model were manipulated
Tags derived from each tool’s schema and the implementation, not from what the tool calls itself. high is the level these add up to.
| Tool | Capability tags | Why the tag was assigned |
|---|---|---|
| No per-tool rows for this release — see the note below. | ||
Toxic-flow graph
The lethal trifecta, checked as a graph rather than as a checklist: untrusted input, a sensitive source and an external sink have to meet before there is a path worth worrying about.
The public result for this release does not print the flow graph, so there is nothing to show here. That is not the same as "no paths were found": what the scan did read is above, under coverage.
Supply chain and provenance
This is the first scan of this surface here, so there is nothing yet to compare it against.
Every result on this tab comes from one deterministic pass over the published package — offline, rule by rule, and auditable line by line above. Same methodology version, same bytes, same score.
Release history
It installs from its repository rather than from a versioned package, so the command cannot be pinned to a release. The history below is for reading; the current release on file is 1.1.1.
- Faster first answer after switching brains, and the full 18-tool surface is listed with the setup steps.
One review per account, from a signed-in account that does not publish this listing. Nothing else is asked — you can say what you think before you install it. Publishers can reply once per review, and a review can be edited or deleted by whoever wrote it.
Any signed-in account that does not publish M1K3 can review it, once — before installing it or after. A review from an account that has installed it is marked as one.
Nobody has reviewed this listing. The rating on the card is the mean of the reviews written here and nothing else, so there is no rating until somebody writes the first — which takes a signed-in account that does not publish it, and nothing else.