Skip to content
Spor

MCP

Spor exposes an MCP (Model Context Protocol) server at /mcp, so an AI assistant — claude.ai, Cowork, Claude Code, or any MCP client — can work with your team’s knowledge graph directly. Connected, the assistant can:

  • Search the graph for what the team already knows about a task, before designing or deciding anything (query_graph).
  • Walk a node’s neighborhood — why it exists, what it depends on, what answered it (get_node, root-mode query_graph, render_lens).
  • Read the ranked decision queue — open work ordered by graph signal, each item with a one-line why (show_queue).
  • Write outcomes back — a decision, a finding, a deferral — as typed, linked nodes the next session inherits (capture, put_node).

Every write is attributed to you. The server stamps the author from the authenticated token, and any author: supplied in a payload is discarded — a connected assistant can never write as someone else.

Prerequisites

Section titled “Prerequisites”
  • A Spor account on a team server. Either your organization’s own Spor server or the hosted product. The MCP surface is a door onto a shared org graph; there is nothing to connect to in purely local mode.
  • A person identity and a personal access token (spor_pat_…), minted for you by a server admin. The connector’s OAuth consent step asks you to paste this token once — see Connecting.
  • An MCP host that supports remote connectors over Streamable HTTP with OAuth 2.1 — claude.ai and Cowork (custom connectors) and Claude Code both qualify. The interactive widget additionally needs a host with MCP Apps support (Claude, Goose, VS Code); other hosts get a text rendering of the same views automatically.

There is no anonymous access: unauthenticated MCP calls are rejected, because every write needs an author.

The tools

Section titled “The tools”

Eighteen tools, framed by the server itself as an ORIENT → TRAVERSE → COMMIT loop rather than a flat list. Full details in the tool reference.

Tool What it does
query_graph Free-text search, or compile one node’s neighborhood (root_id)
get_node One node’s full markdown, edges, and revision
show_queue The ranked decision queue, as data — “what’s next”
render_queue The same queue, with the interactive widget attached
recent_changes What changed in the graph since a commit or a point in time
analytics Created-vs-completed, throughput, cycle time, WIP, bottlenecks
schema Introspect the live schema registry — the contract as data
capture Raw prose in, typed and linked nodes out — the default write door
put_node Create or update one node from full markdown
add_edge Add one typed edge between two nodes
remove_edge Withdraw one edge
set_status Change one node’s status, gated by the schema
set_priority Set or clear the human priority override (p1–p3)
propose_correction Pin, exclude, or add guidance to future briefings
ask_question File a question, routed to whoever stewards the closest node
run_workflow Start a run of an active workflow
render_lens Run a named saved view (board, table, lineage tree)
render_program Progress and gating tree for a workstream root

In this section

Section titled “In this section”
  • Connecting — adding the connector in claude.ai or Claude Code, the OAuth flow you’ll see, and revoking access safely.
  • The operating loop — the ORIENT → TRAVERSE → COMMIT mental model the server teaches connected assistants.
  • Tool reference — every tool: purpose, key parameters, when to reach for it.
  • The widget — the interactive queue, lens, and program views on hosts that support embedded apps.