Skip to main content
bryel runs a remote MCP server so your coding agent can query your own trace data — runs, intents, tokens, cost, tool trajectories — with BQL, right from your editor. It’s OAuth-gated and read-only.

​
Connect

1

Add the MCP server

claude mcp add --transport http bryel https://mcp.bryel.ai/api/mcp
2

Authorize

The first time your agent uses it, a browser opens — sign in to bryel and pick the project the agent may query. That project is bound to the connection; it’s the only one this agent can ever see.
3

Ask away

Your agent now has three tools. Try: “Using bryel, how many traces errored in the last 7 days, and which intents cost the most?”

​
Tools

ToolWhat it does
bryel_schemaLists the queryable fields, types and operators, plus your project’s live intent labels and model names. Call this first.
bryel_validateChecks a BQL filter for errors (with did-you-mean suggestions) without running it.
bryel_queryRuns a BQL filter and returns the match count + a sample (model, status, tokens, cost, steps, tools, truncated input).
Everything is read-only and scoped to the single project you picked at consent — the agent can’t name another project.

​
BQL in 30 seconds

BQL is a filter expression over your traces. Strings use single quotes; combine with and / or / not. 
status = 'error'
intent = 'add_pricing' and cost_usd > 0.5
tool_count >= 10 and repeated_calls > 0
model = 'anthropic/claude-opus-4-7'
Call bryel_schema to see every field, operator, and your project’s actual intent labels + model names.

​
Give your agent the playbook

Paste this into your AGENTS.md / CLAUDE.md (or .cursorrules) so the agent uses bryel well: 
## Querying bryel (observability)

You have bryel MCP tools to query THIS project's LLM/agent traces:

- `bryel_schema` — call FIRST to learn the fields, operators, and the project's
  real intent labels + model names.
- `bryel_validate` — check a BQL filter before running it.
- `bryel_query` — run a BQL filter; returns a match count + sample rows.

BQL is a filter expression (single-quoted strings; combine with and / or / not):
  status = 'error'
  intent = 'add_pricing' and cost_usd > 0.5
  tool_count >= 10 and repeated_calls > 0

Use it to investigate error rates, cost/token outliers, which intents are
expensive, repeated/looping tool calls, and slow runs. Always ground on
`bryel_schema` before writing a query, and `bryel_validate` if unsure.
One connection = one project (chosen at consent), read-only. To query a different project, add a second MCP connection and pick that project. Removing the connection (or losing access to the project) cuts off the agent immediately.