Skip to content

Gemini Code Assist Integration

Pisama connects to Gemini Code Assist via the Pisama MCP server. Gemini Code Assist speaks MCP over stdio — no extra transport or auth surface needed.

Overview

From inside Gemini Code Assist you can:

  • Query traces — find agent runs by framework, status, or date range.
  • Inspect detections — see which failure modes fired and with what confidence.
  • Get fix suggestions — per-detection remediation options with risk ratings.
  • Submit feedback — correct false positives to improve detector recall.
  • Create and run scorers — natural-language quality metrics against your traces.

The Pisama MCP server calls the Pisama backend API (api.pisama.ai) on each tool invocation. An API key and tenant ID are required — get them from pisama.ai/settings/api-keys.

Prerequisites

  • Python 3.11+ accessible from your editor's terminal
  • API key + tenant ID
  • Pisama backend package installed (see below)

Installation

cd ~/pisama/backend
pip install -e ".[mcp]"

Configure Gemini Code Assist

In VS Code or JetBrains with Gemini Code Assist installed, open the MCP servers panel and add:

{
  "mcpServers": {
    "pisama": {
      "command": "python",
      "args": [
        "-m", "app.mcp.server",
        "--base-url",  "https://api.pisama.ai",
        "--api-key",   "pisama_YOUR_KEY",
        "--tenant-id", "YOUR_TENANT_UUID"
      ],
      "cwd": "/path/to/pisama/backend"
    }
  }
}

Reload Gemini Code Assist. Confirm the server is connected by asking:

show me recent pisama detections

You should see a paginated list of detections with types and confidence scores.

Available Tools

Tool Purpose
pisama_query_traces Search traces by framework, status, or date
pisama_get_trace_detail Full trace detail including all agent states and timeline
pisama_query_detections Search detections by type, confidence, or trace
pisama_get_detection_detail Detection detail with explanation and evidence
pisama_get_fix_suggestions Remediation options for a detection
pisama_apply_fix Apply a suggested fix
pisama_submit_feedback Mark a detection as correct or false positive
pisama_create_scorer Create a custom quality scorer from a description
pisama_list_scorers List scorers for this tenant
pisama_run_scorer Run a scorer against recent traces
pisama_evaluate_conversation Full conversation quality analysis for a trace
pisama_generate_source_fix Generate a source code patch for a detection

See the MCP Server guide for full argument reference and troubleshooting.

Example prompts

  • Find failures: "show me pisama detections from the last hour"
  • Investigate: "get the full detail for detection abc123"
  • Remediate: "what fixes are available for detection abc123?"
  • Analyze quality: "run the pisama conversation evaluator on trace xyz789"
  • Custom metric: "create a pisama scorer for tool-use efficiency — penalize redundant tool calls"

Tier behavior

pisama_query_detections returns detections at or above min_confidence (default: no filter). Set min_confidence: 70 to restrict to high-confidence detections matching the production F1 gates in the calibration report.