MCP Server — Limits, Timeouts, and Performance Notes
The Picasi MCP server is designed for fast data access (< 5 seconds). This page lists the current technical limits and timeout rules so that AI assistants can properly scale their requests.
No request allowance
The number of MCP requests is not limited — on no plan. Picasi still counts the calls per Workspace, but only for the statistics; there is no threshold that blocks access.
Tool Limits
Limits per tool call, as currently accepted by the server.
search_updates
| Parameter | Limit |
|---|---|
limit | 1 – 100 (default: 20) |
days | 1 – 365 (default: 7) |
date_from / date_to | Any time period; overrides days |
If limit is set to more than 100, the server truncates the value to 100 and indicates this in the meta block (total_available shows the actual available count).
list_sources
No hard upper limits. The response contains all sources in the Workspace. For very large Workspaces, include_channels: false (default) is recommended to keep the payload small.
list_reports
| Parameter | Limit |
|---|---|
limit | Default 10, no known hard upper limit |
All Other Tools
There are no hard response limits for get_update_details, get_report, get_team_context, list_tags, or the write tools. Payloads are tied to the resource (a report, an update, a team).
Response Times
Data queries are optimized for sub-second response times.
| Tool | Expected Response Time |
|---|---|
search_updates | typically < 1 s |
list_sources | typically < 1 s (without channels) |
list_sources with include_channels: true | 1–3 s (Workspace-dependent) |
get_update_details | typically < 1 s |
get_report | 1–3 s (report-dependent) |
get_team_context | typically < 1 s |
Write tools (create_*, update_*, delete_*, toggle_*) | typically < 2 s |
MCP Timeouts
The Picasi server does not enforce its own server timeout. The timeouts of the respective AI clients apply.
| Client | Effective Client Timeout |
|---|---|
| Claude Code (CLI) | Default 30 s per tool call |
| Claude.ai | Default 30 s per tool call |
| ChatGPT (Custom Connector) | Default 30 s per tool call |
Therefore, analysis tools with long runtimes (generate_feedback, summarize_updates — 15 to 45 seconds) are currently disabled. For longer analyses, use the Picasi app or retrieve the results of a previously generated report via get_report.
Rate Limiting
There is no separate allowance for MCP requests. The general protection mechanisms of the Picasi platform apply.
Querying sparingly still pays off:
- Cache results when identical queries are made repeatedly.
- For bulk evaluations, set
limitexplicitly instead of paginating through small pages in a loop. - For multiple parallel requests, prefer a single tool call per logical query.
Payload Structure
All tool responses provide three blocks:
result– Payload (updates, sources, reports, …)meta– Execution time, applied filters,total_availablefor searcheshints– Suggestions for useful follow-up calls
total_available in meta shows how many results would be available in total, even if they were not returned. This allows the client to decide whether a targeted query is worthwhile.