OloLand Plugins

The verifier stack for buy-side underwriting, as installable agent plugins.

This repo is OloLand's plugin marketplace for Claude Cowork (Claude Desktop), Claude Code (CLI), Codex, and Cursor / Grok Bot. It hosts the OloLand plugins that augment Anthropic's first-party finance plugin lineup (private-equity, financial-analysis, investment-banking, equity-research, wealth-management) and expose the same OloLand MCP server on Cursor-format hosts.

Anthropic ships breadth. OloLand ships depth. Same Cowork session. Deterministic computation. Defensible record.

---

Plugins in this marketplace

Plugin	Status	What it does
ololand-dd	v1.26.2	Institutional due diligence: deterministic financial engines, 311-factor risk taxonomy (67 diligence categories), analytical workbench tools, verified forensic screen workflow, war-game RL strategy simulation, and a flywheel that retrains from analyst corrections.
ololand-forensic-qoe	v0.6.5	Forensic earnings workflows with explicit evidence and data gaps. Individual MCP tools and self-serve PDF reports have different coverage; reports are included in Pro and above. OpenAI submission export and draft review cases are documented in the submission guide.
ololand-compliance-hooks	v0.2.4	Drop-in compliance, citation, and provenance hooks for Anthropic's Claude Cowork finance plugins. PreToolUse MNPI guard, PostToolUse citation enforcer, audit-log writeback. Populates the empty hooks/ scaffold Anthropic's verticals ship with.
cim-generator	v1.1.1	14-section CIM generator with provenance — sell-side memorandums built from your reconciled deal data (financial snapshots, risk insights, market research), not LLM prose. View, edit, and export the finished CIM (PDF, DOCX, PPTX) from the deal workspace.

The plugins compose additively with each other and with Anthropic's first-party finance plugins.

---

Maintainer workflow

This repository generates Claude, Codex, and Cursor / Grok Bot plugin artifacts from canonical YAML metadata. Edit marketplace.yaml and plugins/*/plugin.yaml, then run:

python3 -m pip install -r scripts/requirements.txt
python3 scripts/generate-plugin-artifacts.py
./scripts/check-plugin-artifacts.sh

See docs/plugin-publishing.md for the full release process.

---

Prerequisites

The only thing you need is a free OloLand account. No env var setup, no manual API key creation. The plugins handle authentication via OAuth automatically when you install them.

Two account rails

Rail	Cost	Access
Developer (free)	Sign up at https://app.ololand.ai/signup	Sandbox-only — full MCP tool access against the included sample deal (Paragon Flight School (Sample)). Unlimited calls. No payment, no credit card.
Platform (paid)	Pro $199/mo, Team $499/mo (2 seats included, +$199/seat), Enterprise (custom) (see pricing)	Access to your own deals + cross-deal memory + outcome flywheel + production volumes

You can install the plugins on a free Developer account and explore everything against the sample deal — DCF, LBO, Monte Carlo, forensic QoE, risk taxonomy, war-game, all of it. When you're ready to run on your own data, upgrade to a paid plan.

What happens on first install (no env var setup needed)

1. Add the marketplace and install a plugin (steps in the Install sections below)
2. The plugin defines an ololand MCP connector that points at OloLand's backend
3. On first invocation, your Cowork / Claude Code prompts you to Authorize the connector via OAuth
4. Click Authorize — your browser opens to OloLand's auth page, you sign in, accept the scopes
5. OloLand's backend auto-provisions a scoped agent key, bound to your OAuth client. Cowork / Claude Code stores it for you
6. Subsequent calls use the stored credential. No env var, no manual setup.

That's it. The OAuth flow happens once and the connector is authenticated for that session and future ones. The same connector is what Grok Bot and Cursor authorize after you install a Cursor-format plugin from this marketplace.

Power user / headless setup (optional)

If you're running the plugins in a headless environment (CI, server-side automation, custom MCP client that can't run OAuth), you can bypass OAuth with a manual agent key:

1. Sign in at https://app.ololand.ai
2. Settings → Agent keys → Create new key
3. Set the env var: export OLOLAND_AGENT_KEY=olo_agent_sk_...

The published .mcp.json files are OAuth-native and do not send an Authorization header. For headless use, add your own MCP server config with Authorization: Bearer ${OLOLAND_AGENT_KEY}. If OAuth and a custom header are both configured by your client, OAuth should be treated as the interactive default.

---

Install — Grok Bot and Cursor

Grok Bot installs plugins in the Cursor format. This repo ships a multi-plugin marketplace at .cursor-plugin/marketplace.json. Each plugin reuses its existing skills/, commands/, agents/, assets, and OAuth-native .mcp.json — there is no API key in the repository.

Grok Bot

1. Open the bot's plugin settings, or ask: Install the OloLand plugins from ololand-ai/ololand-plugins
2. Add the GitHub marketplace ololand-ai/ololand-plugins
3. Install ololand-dd for the deal, diligence, and valuation MCP tools (get_deal, run_due_diligence, run_lbo_model, generate_ic_package, and the rest of the OloLand server). Optionally install ololand-forensic-qoe, cim-generator, and ololand-compliance-hooks
4. When prompted, authorize the ololand MCP connector via OAuth. No env var or pasted token is required

Cursor

1. Settings → Plugins → add a marketplace from GitHub: ololand-ai/ololand-plugins
2. Install the plugins you need from the OloLand marketplace
3. Authorize the ololand connector on first use

After install, a plugin detail view should show the shared skills plus the ololand connector. If the connector appears installed but never offers sign-in, remove and re-add the plugin, then authorize again.

Headless Cursor / CI hosts that cannot complete OAuth can add their own MCP config with Authorization: Bearer ${OLOLAND_AGENT_KEY}. The published .mcp.json files stay OAuth-native and do not embed that header.

---

Install — Claude Cowork (Claude Desktop)

The recommended path for analysts running deals through the Cowork UI.

Step 1 — Open the plugin manager

1. Open Claude Desktop
2. Switch to the Cowork tab (top of window)
3. Click Customize in the left sidebar
4. Click Plugins in the Customize submenu

Step 2 — Add the OloLand marketplace

1. Click the + button next to "Personal plugins" (or the equivalent button in the Plugins panel header)
2. Choose Create plugin → Add marketplace
3. In the URL field, enter:

   ololand-ai/ololand-plugins
   

4. Click Sync
5. After ~5-10 seconds, a new pill ololand-plugins appears with three plugin cards: Ololand dd, Ololand forensic qoe, Ololand compliance hooks.

Step 3 — Install the plugins

Click the + button on each card to install. After all three installed, your Personal plugins sidebar should list:

• Ololand dd
• Ololand compliance hooks
• Ololand forensic qoe

Step 4 — Verify

Click Ololand dd in the sidebar. The detail view should show:

Field	Expected value
Source	Marketplace (ololand-plugins)
Version	1.20.4 or higher
Author	OloLand
Skills tab	~18 skills including /playbook-recall, /calibrate-vs-history
Agents tab	3 agents: dd-analyst, forensic-screener, war-game-strategist
Connectors tab	1 connector named ololand

If Agents shows 1 instead of 3, you're on an older version. Remove the marketplace and re-add following Step 2.

Step 5 — Authorize the OloLand connector

The first time you invoke a command that calls OloLand's backend (e.g., /ololand-dd:dd-analyze), Cowork prompts you to authorize the ololand MCP connector. Click Authorize → sign in to OloLand if you haven't yet → accept the scopes. The connector stores the credential for subsequent calls. No env var setup needed.

If you're new to OloLand, signing in here also creates your free Developer account on the spot — you don't have to visit app.ololand.ai/signup separately.

---

Install — Claude Code (CLI)

For developers, IT teams, and anyone running Claude Code locally.

Step 1 — Add the marketplace

In any Claude Code chat session:

/plugin marketplace add ololand-ai/ololand-plugins

Step 2 — Install the plugins

/plugin install ololand-dd@ololand-plugins
/plugin install ololand-forensic-qoe@ololand-plugins
/plugin install ololand-compliance-hooks@ololand-plugins

Step 3 — Reload to activate

/reload-plugins

Step 4 — Verify

/plugin

This opens the plugin manager TUI. Tab to the Installed tab and confirm all three OloLand plugins are listed and enabled. Open ololand-dd detail to verify version 1.20.4 and the three agents.

Step 5 — Authorize the OloLand connector

The first time you invoke a command that calls OloLand's backend (e.g., /ololand-dd:dd-analyze), Claude Code triggers the OAuth flow for the ololand MCP connector. Your browser opens to OloLand's auth page → sign in → accept the scopes. Claude Code stores the credential. Subsequent calls authenticate automatically.

If you're new to OloLand, the auth page lets you sign up on the spot — no separate trip to app.ololand.ai/signup.

For headless / CI use only: set OLOLAND_AGENT_KEY=olo_agent_sk_... in your environment to bypass the OAuth flow. Get the key at app.ololand.ai → Settings → Agent keys → Create new key. OAuth takes precedence if both are configured.

---

Usage — ololand-dd

The flagship due-diligence plugin. Skills are auto-fired when relevant; commands are explicit invocations.

Commands

Type / in any chat to see autocomplete. Plugin commands are namespaced as /ololand-dd:<command>.

Command	What it does
/dd-analyze <deal_id>	Run the deal extraction and risk-analysis pipeline; valuation reads or fresh model candidates are separate explicit /valuation actions
/new-deal <ticker_or_name>	Create a new deal from a ticker (auto-fetches 10-K) or company name
/risk-matrix <deal_id>	Render the risk taxonomy (67 categories / 311 risk factors) as an interactive tile
/risk-report <deal_id>	Structured risk report — severity × likelihood × velocity, with $-impact
/valuation <deal_id> read [method]	Read governed DCF / LBO and source-backed Comps. To create a fresh model candidate, explicitly use /valuation <deal_id> run <method> or /valuation <deal_id> refresh <method>; bare and legacy method-only calls are read-only.
`/qoe-analysis <deal_id> [latest	run]`
/verify <deal_id>	Defend every number: verify an OloLand artifact against source docs + engine runs, or grade an externally-drafted memo (e.g. from Claude Cowork) per-claim against the deal corpus
`/compliance-analysis <deal_id> [ofac	hsr
`/scenario-analysis <deal_id> [stress	market
/earnings-analysis <deal_id>	Analyze earnings-call transcript segments for diligence signals
`/verification-marketplace <deal_id> [status	list
`/ic-package <deal_id> [status	generate
`/value-impact <deal_id	company
`/deal-health <deal_id> [actions	create-action]`
`/workbook [list	get
/company-discovery <query>	Search the company-discovery surface from natural-language or structured criteria
`/watchlist [list	create
`/advisory <deal_id> [list	request]`
`/financing <deal_id> [status	analyze
/managed-context-agent <deal_id> <agent_key>	Launch a configured managed context agent when enabled
/okf-export <deal_id>	Export a machine-readable OKF deal bundle
/war-game <deal_id> [scenarios]	16-quarter MaskablePPO competitive simulation across scenarios
/similar-deals <deal_id>	Cross-deal memory: find similar past deals, accuracy patterns, valuation ranges
/playbook-recall <deal_id>	What worked, what didn't, what was missed in similar past deals from your firm's history
/calibrate-vs-history <deal_id>	Apply your firm's historical projection bias to recalibrate the deal's headline numbers
/talk-to-deal <deal_id>	Conversational session: ask anything about the deal, voice-optimized output
/deal-search <deal_id>	Full-text + semantic search across a deal's data room
/unit-economics <deal_id>	SaaS cohort retention (NDR/GRR), LTV/CAC, payback, magic number, rule of 40
/source <criteria>	Source new M&A/PE deals from criteria; enrich with contact data; dedupe vs CRM

Sub-agents

In Cowork, click an agent in the Agents tab and it loads as the lead agent for the conversation. In Claude Code CLI, agents auto-fire based on context, or invoke explicitly with /agents and pick one.

Agent	Use when
dd-analyst	Full DD workflows, IC memo prep, deep deal investigation. The general-purpose specialist.
forensic-screener	Pre-LOI forensic deep-dive. Runs all 7 forensic primitives, reconciles findings against source hierarchy, produces an IC-defensible exclusion schedule with severity, $-impact, and citations.
war-game-strategist	Stress-testing the deal thesis against competitor responses. Runs 4-scenario MaskablePPO simulation, calibrates vs. similar-deal outcomes, overlays on deterministic DCF, recommends bid adjustment.

Example workflow — full Pre-LOI DD

In Cowork (or Claude Code), with a deal in your OloLand account:

/ololand-dd:new-deal ACME-Manufacturing

↓ creates the deal

/ololand-dd:dd-analyze deal_acme_2026

↓ runs the full multi-agent DD

Then for forensic depth:

/ololand-dd:risk-matrix deal_acme_2026

↓ renders the risk-taxonomy tile

@forensic-screener run a Pre-LOI screen on deal_acme_2026

↓ delegates to the forensic specialist sub-agent

For thesis stress-test:

@war-game-strategist test the buyout thesis on deal_acme_2026 across all 4 scenarios

---

Usage — ololand-forensic-qoe

The standalone Pre-LOI Forensic Screen wedge. Use when you want forensic depth without the full DD plugin.

Commands

Command	What it does
/forensic-screen <deal_id>	Full battery: Beneish + Benford + EBITDA bridge + journal-entry + lapping + working-capital + revenue-quality. Returns severity-scored exclusion schedule.
/beneish <deal_id>	Beneish M-Score (private-company adjusted). 8-variable earnings-manipulation probability.
/benford <deal_id>	Benford's Law first-digit testing on GL transactions. χ² + MAD.
/ebitda-bridge <deal_id>	EBITDA bridge with adjustment classifier. Tags every add-back as accepted / recurring / questionable.
/journal-test <deal_id>	Journal-entry anomaly testing. Period-end concentration, round-numbers, posting authority, weekend entries.
/lapping-check <deal_id>	AR-lapping fraud detection. Customer-to-cash trace, application gap detection.
/conflicts <deal_id>	Cross-document conflict detector for mismatched figures, dates, ownership, terms, and obligations.
/verify <deal_id>	Verify an externally drafted artifact against the ingested deal corpus.

Required inputs

• Beneish: 2+ years of income statement + balance sheet data
• Benford: GL export with ≥1,000 line items
• EBITDA bridge: management adjustment schedule + 2-3 years historicals
• Journal-entry tests: GL with timestamps + posting user IDs
• Lapping: AR aging + cash receipts journal with customer-level detail
• Full screen: skip-and-report behavior — runs every primitive whose inputs are present, reports which were skipped and why

Example workflow — $50M EBITDA target screen

/ololand-forensic-qoe:forensic-screen deal_target_2026

Returns in 60-90s an IC-defensible exclusion schedule with:

• Reported EBITDA $X.XM → defensible EBITDA $Y.YM (delta $Z.ZM)
• Each forensic primitive's verdict (pass / warning / fail)
• Adjustments table with $ impact and source citations
• Severity-scored findings (low / medium / high / critical)
• Recommendation: proceed / proceed with caveats / kill

---

Usage — ololand-compliance-hooks

The drop-in hooks plugin. Most users don't invoke this directly — it runs automatically once installed.

What it does (passive)

Hook	Trigger	Action
PreToolUse: MNPI guard	Bash, Write, Edit, MultiEdit	Blocks tool calls referencing material non-public information patterns ("material non-public", "insider trading window", etc.) unless input contains # mnpi:cleared marker
PostToolUse: citation enforcer	After CIM / IC memo / dossier generation	Scans for $-amounts, %, multiples without source citation. Warns by default.
PostToolUse: provenance writeback	After every generative output	Appends NDJSON to ~/.ololand/provenance/YYYY-MM-DD.ndjson + optional API mirror
Pre + PostToolUse: audit log	Every mcp__ololand__* call	Mirrors invocations to ~/.ololand/audit/YYYY-MM-DD.ndjson
SessionStart: banner	New session	Confirms hooks armed

Configuration via env vars

Variable	Default	Purpose
OLOLAND_CITATION_BLOCK	0	Set to 1 to upgrade citation enforcer from warn to deny
OLOLAND_AGENT_KEY	unset	Mirror provenance + audit lines to OloLand's API
OLOLAND_API_URL	https://app.ololand.ai	Override for self-hosted OloLand deployments

Local logs

Inspect provenance + audit ledgers any time:

jq -c '.' ~/.ololand/provenance/$(date -u +%Y-%m-%d).ndjson
jq -c '.' ~/.ololand/audit/$(date -u +%Y-%m-%d).ndjson

These ledgers are append-only NDJSON. Rotate or compress per-day as needed.

---

Troubleshooting

Cowork: "Marketplace sync failed. Check the repository URL and try again."

Most common cause: cached marketplace from a prior install attempt. Click ... on the failing pill → Remove. Then re-add.

If still failing, fall back to the Upload plugin path (zip upload). Per Cowork issue #38429, source: "manual" plugins are protected from the auto-removal bug that affects source: "github" plugins.

Cowork: plugin disappears after restart

Known issue #38429 and #39274. RemotePluginManager removes 3rd-party github-sourced plugins on restart. Workarounds:

• Reinstall via "Upload plugin" (zip path) instead of "Add marketplace"
• Or wait for Anthropic to ship the fix

Cowork: chat says "Unknown skill: plugin"

Cowork chat does NOT support /plugin slash commands. Plugin management is UI-only:

• Customize → Plugins → Browse / Add marketplace / Upload plugin

The /plugin command set works in Claude Code CLI only.

Cowork: showing old version (1.4.x) instead of 1.5.x

Likely cached install from before the marketplace rename (was ololand-dd-plugin, now ololand-plugins). Remove the old marketplace pill, add ololand-ai/ololand-plugins fresh, install.

Verify v1.20.4 by clicking the plugin → Agents tab should show 3 agents (dd-analyst, forensic-screener, war-game-strategist).

CLI: /plugin install fails with "marketplace not found"

/plugin marketplace update ololand-plugins

Or if that fails:

/plugin marketplace remove ololand-plugins
/plugin marketplace add ololand-ai/ololand-plugins

Tools fail with auth errors ("OLOLAND_AGENT_KEY not set", "401 Unauthorized", etc.)

Most common cause: you skipped or canceled the OAuth flow on first invocation.

Fix in Cowork: Customize → Connectors → click the ololand connector → click Reconnect (or Disconnect then click the connector again to retrigger OAuth). Sign in to OloLand and accept scopes.

Fix in Claude Code CLI: run any /ololand-dd: command — Claude Code retriggers the OAuth flow when it detects an unauthenticated connector. Sign in in the browser when prompted.

Headless / CI fallback: if OAuth isn't possible (CI, scripts), generate an agent key at app.ololand.ai → Settings → Agent keys, then export OLOLAND_AGENT_KEY=olo_agent_sk_... in your environment.

Plugin skills not appearing after install

Run:

/reload-plugins

If still missing, clear the plugin cache:

rm -rf ~/.claude/plugins/cache

Then re-install. Restart Claude Code / Claude Desktop after the cache clear.

---

Why this exists

OloLand's thesis: Anthropic's finance plugins are "reference templates — they get better when you tune them to how your firm works" (their own README). That is exactly what we are: the institutional tuning.

Anthropic ships breadth — five vertical plugins, ten named agents, eleven read-only data connectors. We ship depth — deterministic computation, structured taxonomy, persistent memory, and the compliance hooks Anthropic's first-party plugins ship empty.

For the full positioning, see the comparison page.

---

Contributing

Each plugin lives under plugins/<name>/ with Claude, Codex, and Cursor manifests that share the same skills, commands, and MCP config. The Claude marketplace catalog at .claude-plugin/marketplace.json and the Cursor marketplace at .cursor-plugin/marketplace.json are generated from marketplace.yaml and must stay in version lockstep with each plugin's plugin.yaml.

Branch from origin/main, open a PR into main (operator ruling 2026-07-14: this repo is main-only — it has no staging environment, so unlike ololand-ai/olo5 there is no staging branch in the flow; the old staging branch is retired and stale). The repo allows merge commits only (rebase/squash disabled).

To bump a plugin version, update both the plugin's plugin.json AND the marketplace.json entry — scripts/check-version-sync.sh (run pre-commit) will fail the commit if they drift.

Support

• Email: support@ololand.ai
• Issues: https://github.com/ololand-ai/ololand-plugins/issues
• Docs: https://ololand.ai

License

Apache-2.0. See LICENSE.