MCP Prompts
cks-mcp also exposes four MCP Prompts — ready-made workflow templates
a client can list (prompts/list) and fill in (prompts/get) to get a
concrete instruction back, without the user having to know which tools to
chain together or in what order.
A prompt doesn't call any tools itself: prompts/get returns a single
user-role text message, built from the prompt's arguments, that the
connected model then acts on normally — using whichever of the 24 tools
that instruction implies.
Available prompts
| Name | Arguments | What it produces |
|---|---|---|
create_knowledge_graph |
topic (required) |
An instruction to build a validated knowledge graph (≥5 objects, ≥3 relations) about topic, then validate it. |
verify_claim |
claim (required), url (required) |
An instruction to verify claim against url and produce a signed VerificationRecord. |
explore_subgraph |
session_id (required), seed_id (required), depth (optional, default 1) |
An instruction to query the subgraph around seed_id in session_id at the given depth. |
branch_and_merge |
session_id (required) |
An instruction to branch session_id, evolve the original and the branch differently, then merge the branch back. |
Example
prompts/get with name: "verify_claim" and
arguments: {"claim": "Dark matter consists of WIMPs", "url": "https://example.org/paper"}
returns:
{
"messages": [
{
"role": "user",
"content": {
"type": "text",
"text": "Use cks-mcp to verify this claim: \"Dark matter consists of WIMPs\". Check the source at https://example.org/paper and create a signed VerificationRecord."
}
}
]
}
The model receiving that message would typically call verify_source,
then merge the resulting VerificationRecord into the relevant session via
evolve_knowledge — but the prompt itself only supplies the instruction,
not the tool sequence; that's still up to the model.
Why these four
Each prompt corresponds to one of the workflows Getting Started
and Tools Reference walk through by hand:
knowledge construction, provenance verification, graph exploration, and
branching — a template for the four most common "which tools do I even
call for this" starting points. If you find yourself typing the same
multi-tool instruction repeatedly, a fifth prompt in PROMPTS (in
src/cks_mcp/prompts.py) is the natural place to save it.