Hub · Connectors Integrations / Cursor · Composer · Agent · Chat

Knovya in your IDE — the decisions follow.

Cursor's Composer is brilliant at any single moment. Then the session ends, and the reasoning ends with it. Knovya is the part that stays — the architecture decision you wrote in October, the postmortem from last sprint, the thread your team pinned about caching strategy. Cursor reaches into the same archive every time you open a file.

Composer reads from your archive. Agent mode cites your ADRs. The decisions your team made last quarter become context the IDE already has, instead of context you copy-paste from another tab.

34 tools · 5 families
~2 min to install
Open MCP standard · OAuth 2.1 + PKCE
The install moment

Two minutes. One config block.

Cursor speaks MCP — the open standard for AI tools. Adding Knovya is one paste into ~/.cursor/mcp.json and a restart.

Path I

Global config

Edit ~/.cursor/mcp.json. Knovya appears in every project you open from this machine.

Path II

Per-project

Drop .cursor/mcp.json at the repo root. Commit it. Knovya scopes to that codebase only — clean for client work.

Path III

One-click

From our connectors page, hit Connect to Cursor. We pre-fill the config; you OAuth in.

Behind the door

Same brain, 34 tools, organized for the IDE.

Composer reaches into Discovery first when you Cmd+K. Read tools fire when you @knovya a doc. Agent mode unlocks Mesh — when more than one AI is in the room.

  • 0 Discovery 7 tools

    Where am I, what’s here. Read-only orientation tools Cursor uses first to map your archive before anything else.

    pingworkspacehomepersonacontextsearchschema
  • 1 Read 7 tools

    Open the note. Trace a link. Pull an attachment. Everything below the search bar — the tools Composer reaches when you cite an ADR or a retro.

    readexperiencememoryfoldershistorylinksattachments
  • 2 Write 4 tools

    Capture, edit, organize, archive. Agent mode surfaces a confirmation step before changing your archive — destructive ops are gated by user consent.

    writeeditorganizedelete
  • 3 Mesh 11 tools

    When more than one AI is in the room. Agent mode in Cursor talks to Claude in another window — presence, channels, shared attention, consensus voting.

    presencechannelscoordinateattentionthoughtscommitteeagentseventsmesh_adminpipelinenotifications
  • 5 Transform 5 tools

    Shape the archive. AI rewrites. Templates. Imports, exports, shares. Where Knovya itself becomes a tool Cursor’s agent can wield.

    aitemplatesexportimportshare
The problem

Cursor's Composer is brilliant in any single session. Then the session ends, and what you decided ends with it.

Yesterday you reasoned through the auth migration. You weighed three approaches, picked one, wrote a long Composer thread justifying the choice. Today you open the file again. Composer starts over — same questions, same trade-offs, no memory of what your team already settled.

The intelligence is enormous. The memory is somewhere else — in a Notion page nobody opens, a Slack thread that scrolled away, a half-finished doc in Google Drive. Composer re-derives it from scratch, every time.

Knovya is the part that doesn’t reset. Same archive, every conversation. The reasoning sticks.

↳ Lineage How we got here, in five steps

Lineage

How the IDE got a memory.

Five turns. Four years. The doorway between Cursor and your archive didn’t exist until last.

  1. 2022

    ChatGPT lands in public.

    AI starts pair-programming inside the chat window. The IDE is still elsewhere — two tabs, copy-paste in between.

  2. 2023

    Cursor launches as the first AI-native editor.

    The chat moves into the IDE. Composer ships. Agent mode arrives the year after. The model finally lives where the code does.

  3. 2024

    Anthropic ships the Model Context Protocol.

    An open spec for AI ↔ tools. Any client can talk to any server, the same way browsers talk to web servers. Knowledge becomes portable.

  4. 2025

    Cursor adopts MCP.

    Composer and Agent mode become MCP clients. Filesystem, GitHub, Context7 — every server in the ecosystem — lights up inside the IDE.

  5. 2026 · today

    Knovya ships the first knowledge-graph MCP for Cursor.

    Filesystem reads files. GitHub reads commits. Knovya reads your reasoning — ADRs, retros, decisions, customer notes. Cursor stops asking the same questions.

First mover

Other Cursor MCPs do one thing well. Knovya does the one thing they don’t.

Context7

Fetches up-to-date documentation for libraries and frameworks.

Docs lookup

GitHub MCP

PR reviews, issue triage, commit operations from inside Composer.

Repo ops

Filesystem MCP

Reads and writes local files outside the active workspace.

Local files

Sequential Thinking

Chain-of-thought scaffolding for multi-step reasoning.

CoT scaffold

Playwright MCP

Browser automation — Cursor drives a real Chromium.

Browser auto

Knovya

Your knowledge graph — decisions, retros, ADRs, customer notes — as first-class IDE context. The reasoning that doesn’t live in the repo.

Decision memory

They stack. Most Cursor power users run two or three MCPs at once — Filesystem for local reads, Context7 for fresh docs, Knovya for the reasoning. Different layers of the same context.

Surfaces

Four moments. One archive.

Where the connector actually shows up — Composer, Agent, the JSON config, the sidebar.

Composer · Cmd+K i.

@knovya   "auth strategy"

› refactor /api/auth using @knovya "auth strategy"
↳ knovya_search returned 3 notes ranked by NoteRank
ADR-014 · Move from JWT cookies to session tokens NR 0.91
Sprint 47 retro · auth migration postmortem NR 0.78
Customer call · SSO requirements from Acme NR 0.64

Composer treats Knovya like any source. Type @knovya and search results stream in, ranked.

Agent mode ii.

Tool calls, cited inline

// Agent reasoning, condensed
Found a stale auth handler. Checking precedents…
knovya_experience(topic="auth migration")
↳ 3 successful precedents · 1 cautionary (Sprint 41)
Sprint 41 hit a token-expiry race. Reading the postmortem…
knovya_read(note_id="adr-014")
✓ Patch drafted with the race condition guarded.

Agent mode picks the right tool per step — Discovery, then Read, then writes back the patch. Each call shows up in the chat as it fires.

~/.cursor/mcp.json iii.

The whole config, one paste.

{
  "mcpServers": {
    "knovya": {
      "url": "https://mcp.knovya.com/mcp",
      "transport": "http"
    },
    "github": { "url": "https://api.githubcopilot.com/mcp/" },
    "context7": { "command": "npx", "args": ["@upstash/context7-mcp"] }
  }
}

OAuth handles sign-in on first call — no API key in the file, no token in your shell history. Restart Cursor and the tool count appears in Settings → MCP.

Cursor · Settings → MCP iv.

Sidebar, once it’s connected.

MCP servers 3 live
github 23
filesystem 9
knovya 34

Each server lists its tool count. Click in to see the manifest — what’s safe, what’s destructive, what scopes the agent has. Audit-friendly by default.

Free 50 MCP calls / month — no credit card
Pro · $15/mo 5,000 calls · encrypted notes
Team · $25/seat Unlimited · shared archive
See pricing
Questions, in advance

Cursor & Knovya — before you ask.

Does Cursor have its own MCP?
Cursor is the MCP client; Knovya is one of the MCP servers it can talk to. Cursor speaks the open Model Context Protocol — a standard maintained by Anthropic and adopted across Claude, ChatGPT, Gemini, Copilot, and Windsurf. Knovya speaks the same protocol, so adding Knovya to Cursor takes a single config block.
How do I add Knovya as an MCP server in Cursor?

Open ~/.cursor/mcp.json (global) or .cursor/mcp.json (per-project). Add a knovya entry with the MCP URL — the snippet we show above is the whole thing. Restart Cursor. The tool count appears in Settings → MCP.

OAuth 2.1 with PKCE handles the sign-in on first call — no API key in the config, no token in your shell history. Most setups take under two minutes.

What's the best MCP server for Cursor?

It depends on what's missing. For documentation lookup, Context7. For repo and pull request operations, GitHub MCP. For local file ops outside the workspace, Filesystem. Knovya is the only MCP server that brings your knowledge graph — decisions, retros, ADRs, customer notes — into Cursor as first-class context.

They stack. Most Cursor power users run two or three MCPs at once; Knovya is the layer that holds your team's reasoning.

Can Cursor's Composer read from Knovya?
Yes. Composer treats Knovya like any other MCP source — once connected, you can call knovya_search, knovya_read, knovya_experience, and twenty more tools directly from a prompt. Agent mode invokes them automatically when the model decides it needs context.
Is Knovya free for Cursor users?
Yes. Knovya Free includes 50 MCP calls per month — enough to try the full memory layer with no credit card. Pro lifts the cap to 5,000 calls for $15/month. Team makes it unlimited at $25 per seat.
How is Knovya different from Context7 or Filesystem MCP?

Context7 indexes public package documentation. Filesystem reads local files. Knovya is your private knowledge graph — the decisions, the sprint retros, the customer interviews, the architecture you wrote.

Different layers of context. Most Cursor power users run two or three MCPs at once; Knovya is the layer that holds your team's reasoning.

Does Knovya MCP work with Cursor's Agent mode?
Yes. Agent mode reads from MCP sources autonomously — it sees Knovya's tools, picks the right one for each step, and cites results inline. Discovery and Read tools are safe by default; Write tools surface a confirmation step before changing your archive.

Cursor in your hands.
Knovya in the room.

Open Composer in the morning, push from Agent mode at night. The code is yours. The reasoning is shared. The doorway closes behind you and opens at the next session.

Or jump to the install snippet, then see pricing.

Open MCP standard · OAuth 2.1 + PKCE. Free tier. 14-day Pro trial. No credit card to start.