How to set it up
How MCP works in Oodle
Connect once. State your intent. Let the agent do the rest, on a loop.
1 · Connect
One command to connect your harness or your agents.
Claude Code, Cursor, Codex, Claude Desktop, Windsurf, or anything else that speaks MCP. Sign in with OAuth or use an API key.
$ claude mcp add --transport http --scope user oodle-ai \
https://us1.oodle.ai/v1/api/instance/acme/mcp \
--header "X-API-KEY: …" # or sign in with OAuth
✓ connected · tools available:
- query_genai_traces
- get_genai_trace
- get_genai_agent_graph
- genai_recommendations
- manage_genai_evaluators
- manage_genai_prompts
Same server for Cursor, Codex, Claude Desktop, Windsurf. Copy the exact command from Settings → Integrations → MCP.
2 · Ask
An MCP tuned for agents. State your intent, it does the rest.
Tools show up when the conversation needs them, not all at once, so the agent’s context stays small. The tools do the correlating on Oodle’s side, across traces, sessions and deploys, so the agent gets an answer rather than raw data to sift. And they cover everything you can do in Oodle, so there is nothing the agent has to come back to you for.
› Why did refund calls start failing yesterday afternoon?
loading skill agent-debugging · 3 tools added for this question
- query_genai_traces
- get_genai_agent_graph
- query_trace_anomalies
initiateReturn returns 503 in 41 sessions since 14:02, right after deploy 8f3c2a of returns-service. 38 of the 41 callers gave up before the agent recovered.
3 tool calls · traces, deploys and sessions correlated on Oodle's side, not in the agent's context
3 · Scale
The performance your agents need.
Agents don’t ask one question at a time. They fire off dozens at once. Each Oodle query fans out across serverless workers, and many queries run side by side without waiting on a shared cluster. Nothing is running between questions, so all that parallel work does not turn into a standing bill.
Why Oodle is fast
- Maximum concurrency: thousands of serverless invocations per query.
How Oodle scales on demand
- Serverless compute scales up for each query, then shuts down when the work is done.
Why Oodle is fast
Maximum concurrency: thousands of serverless invocations per query.
4 · Loop
Let agents find the issue and raise the fix. Then run them in a loop.
Point an agent at Oodle on a schedule. Each run it looks for new issues, reads the sessions behind them, works out the cause and opens a pull request with the fix. Nothing new? It says so and waits for the next run.
- 09:00initiateReturn 503 in 41 sessionsPR #412 opened · retry on 503 with backoff
- 09:30no new issueswaiting for the next run
- 10:00checkInventory timing out for EU usersPR #418 opened · raise timeout, add fallback
- 10:30#412 merged · 503s gone from the last 200 sessionsissue closed