# Converra

> Converra is an MCP server for ai models and llm tooling. Find, fix and verify production AI agent failures from conversation traces. It runs as a hosted remote server, so there is nothing to install locally.

- **Kind:** MCP server
- **Platforms:** API, Web
- **Website:** https://converra.ai/docs/mcp/coding-agents
- **Repository:** https://github.com/converra/converra-agent-kit
- **MCP Registry name:** io.github.converra/converra
- **Categories:** [AI models and LLM tooling](https://ai-powered.com/categories/ai-and-llm), [developer tools](https://ai-powered.com/categories/developer-tools)
- **Listing updated:** 2026-11-15

## About

Converra reads production conversations, diagnoses each failure at the step where it happened, generates a bounded prompt or configuration fix, and tests it head to head against the current agent in simulation. Customers approve what ships, and Converra reports a verdict on the targeted failure afterwards.

Verdicts follow stated thresholds. A fix is verified after at least 15 post-change conversations with no recurrence, improving if recurrences fell, not fixed if they did not, regressed if the overall failure rate rose, and confounded if the prompt was edited elsewhere. It waits for at least 5 diagnosed conversations first.

It reads traces from LangSmith, Langfuse, its SDKs or the API. Claude Code and Codex connect over OAuth, Cursor with an API key. The docs say not to submit system prompts as conversations.

## Features

- **Step-level diagnosis:** Finds the failing step and root cause.
- **Bounded fixes:** Prompt or configuration changes.
- **Simulation tests:** Head-to-head, regression and benchmarks.
- **Production verdicts:** Verified, not fixed, confounded.
- **Trace sources:** LangSmith, Langfuse, SDKs, API.
- **Repo rules:** Instructions for AGENTS.md.

## Good for

- Find why an agent regressed
- Test a prompt change
- Check a fix on live traffic
- Add transcripts

## Install

### Claude Code (Terminal)

```sh
claude mcp add --transport http converra https://converra.ai/api/mcp
```

### Claude Desktop (Settings → Connectors → Add custom connector)

```sh
https://converra.ai/api/mcp
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "converra": {
      "url": "https://converra.ai/api/mcp"
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "converra": {
      "type": "http",
      "url": "https://converra.ai/api/mcp"
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "converra": {
      "serverUrl": "https://converra.ai/api/mcp"
    }
  }
}
```

### Grok Bot (Grok Bot desktop app → Connect apps → Add a custom connector)

```sh
https://converra.ai/api/mcp
```

Enter a Name (for example "Converra"), choose Type: Remote HTTPS, paste this as the Server URL and choose Add plugin.

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "converra": {
      "type": "http",
      "url": "https://converra.ai/api/mcp"
    }
  }
}
```

## Tools

- `get_account`: Starts a session by returning account details.
- `list_agents`: Lists monitored agents.
- `get_agent_status`: Shows an agent's status.
- `get_insights`: Shows insights on failures.
- `get_run_step_diagnosis`: Shows the failing step, mode and root-cause layer.
- `get_worst_run`: Finds the worst run.
- `fix_worst_run`: Turns a recurring failure into a fix.
- `trigger_optimization`: Starts an optimization.
- `get_optimization_details`: Shows optimization details.
- `get_variant_details`: Shows variant details.
- `apply_variant`: Applies a variant.
- `simulate_agent`: Simulates an agent.
- `regression_test`: Runs a regression suite.
- `simulate_ab_test`: Runs a head-to-head test.
- `benchmark_models`: Benchmarks models.
- `get_deployment_impact`: Reports the production verdict.
- `get_verification_evidence`: Shows verification evidence.
- `create_run`: Adds a real conversation.
- `get_run_insights`: Shows insights for a run.

## FAQ

### How is a fix verified?

15+ post-change conversations and no recurrence.

### Is it an observability tool?

No, it reads traces and acts on them.

### How do coding agents connect?

OAuth in Claude Code and Codex, API key in Cursor.

### What should create_run take?

Real user and assistant turns only.

### How do I add Converra to Claude Code?

Run this command in your terminal: claude mcp add --transport http converra https://converra.ai/api/mcp

### What are the best alternatives to Converra?

The closest alternatives to Converra are conductor, VerifyAX, Concord MCP, Agent evidence and agent-config.

## Where this comes from

- Description: written from the project's README and the project's website.
- Install steps: from the official MCP Registry entry.

## Alternatives

- [conductor](https://ai-powered.com/mcp/conductor): conductor is an open-source MCP server for developer tools.
- [VerifyAX](https://ai-powered.com/mcp/verifyax): VerifyAX is an MCP server for ai models and llm tooling.
- [Concord MCP](https://ai-powered.com/mcp/concord-mcp): Concord MCP is a freemium MCP server for developer tools.
- [Agent evidence](https://ai-powered.com/mcp/agent-evidence): Agent evidence is an open-source MCP server for ai models and llm tooling.
- [agent-config](https://ai-powered.com/mcp/agent-config): agent-config is an open-source MCP server for developer tools.
- [Convocore](https://ai-powered.com/mcp/convocore): Convocore is a freemium MCP server for communication.
- [Argentic](https://ai-powered.com/mcp/argentic): Argentic is a freemium MCP server for ai models and llm tooling.
- [TriedAgents](https://ai-powered.com/mcp/triedagents): TriedAgents is a free MCP server for ai models and llm tooling.

---
Source: https://ai-powered.com/mcp/converra
