Reference
MCP plans, coins and limits
Which plans include the OClever MCP server, what each action costs in coins, rate limits and how your data is handled.
Plans
| Plan | What the MCP server can do |
|---|---|
| Starter | All read tools, resources and prompts. |
| Growth | Everything in Starter, plus write tools (prompts, competitors, recommendations) and coin actions (collection runs, article drafts). |
| Scale | Everything in Growth, plus Prompt Lab, Agent Readiness tests and the agent readiness report. |
| Enterprise and custom plans | Everything. |
MCP uses your workspace's existing plan: there is no separate MCP add-on. If a tool needs a higher plan, it returns a clear message saying which plan unlocks it instead of failing silently.
Coin costs
Read tools never cost coins. Actions cost exactly what the same action costs in the dashboard and come out of the same balance. Every coin tool previews its cost first.
| Action | Tool | Coins |
|---|---|---|
| Prompt Lab question | run_prompt_lab | 10 per engine asked (1 to 3 engines) |
| Agent Readiness test | start_agent_readiness_test | 2 per task; a flat 50 when competitors are included |
| Article draft | draft_article | 40 per article, plus the cost of any images |
| Collection run | run_collection_now | 1 per prompt per engine |
| Create prompts, competitors, recommendation updates | create_prompts | Free (counts against plan limits) |
If a run fails on OClever's side, its coins are refunded automatically, just as in the dashboard. get_coin_balance shows your balance and what was spent, including through MCP.
Rate limits
- 120 calls a minute per user, across all your access tokens and connected apps.
- 10 coin-spending calls a minute per user.
- No monthly cap. Use the server as much as you like within the per-minute limits.
- Long jobs (Prompt Lab, Agent Readiness, article drafts) return a run id straight away; the assistant polls with the matching get_ tool.
- Results are paged at up to 100 rows so they fit in an assistant's context.
Going over a limit returns a rate_limited error that tells the assistant how long to wait.
How your data is handled
- The assistant sees only what the tools return, for the workspaces you approved. Your OClever password is never shared with it.
- Data returned to your assistant is then handled by that assistant's provider under your agreement with them. Choose the workspaces you connect with that in mind.
- OClever logs each call (tool name, connection, workspace, time, duration, coins and whether it succeeded) so you can see activity in Settings → AI assistants. We store a fingerprint of the arguments, not the arguments themselves.
- Access tokens and OAuth tokens are stored as one-way hashes. A leaked database copy cannot be used to call the server.
- Text collected from AI engines is returned as labelled third-party content, and the server tells the assistant not to follow instructions inside it.
See the privacy policy for the full picture, and revoking access to disconnect at any time.
Read-only connections
To make sure an assistant can never change anything, connect it with the read-only URL. Write and coin tools are then not listed at all, regardless of the token's permissions.
Read-only server URLhttps://mcp.oclever.com/mcp/readonly