How can we help?
Start with the product overview, connect your coding agent, or find a focused guide. The Help button in the header remains available for tours and search.
Guides
Understand what Brain Researcher does
See how Brain Researcher connects questions, evidence, workflows, execution handoffs, and human review without replacing scientific judgment.
Connect Brain Researcher to your coding agent (MCP setup)
4-step flow: generate a brk_ token, paste the config into Claude Code/Codex/Cursor, verify with server_info + system_self_test, then hand off a workflow.
Explore graph-backed evidence in the Knowledge Graph
Search datasets and nodes and inspect KG evidence across task, disease, and ONVOC views before committing to a run.
Browse the worked case reports
Open the demo catalog of public use-case reports with replay evidence, artifact packages, and MCP handoff context.
NeuroMark schizophrenia connectivity across analysis choices
Inspect the Figure 4A-C case on estimator-dependent schizophrenia connectivity and its corrected human audit.
Cocaine-use-disorder connectivity robustness
Inspect the Figure 4D-E multiverse in which none of five prespecified findings survived correction.
Cross-cultural social cognition meta-analysis
Inspect the Figure 4F exploratory meta-analysis and the evidence limits created by the small, imbalanced literature.
HCP prediction workflow search
Follow the Figure 5 trajectory from 116 workflow evaluations to a frozen matched comparison.
TRIBE speech-tools representation geometry
Follow the Figure 6 trajectory from an open sound screen through new panels and collections.
Additional use case: Connectome hubness and decoding
Inspect a separate curated report on hubness, decoding, and generative fidelity outside the five manuscript cases.
Read the documentation
Start from the docs landing page for the Help Center, resources, and platform overview.
Browse the workflow library
Discover reusable workflows and pipelines you can plan once and run anywhere via your coding agent.
Quick answers
How do I get an MCP token?
Go to /mcp/setup and sign in, then under 'Personal MCP token' click Generate token (or Rotate token if you already have one). The secret is shown once - copy it immediately. There is one active token per user, so generating a new one rotates and immediately revokes the previous one. The token format is brk_<kid>.<secret>. Store the raw token in BR_MCP_TOKEN in an environment your client inherits, such as a user shell profile.
My agent can't see the MCP tools. How do I fix it?
Most often the token is wrong. Use the raw brk_<kid>.<secret> value in BR_MCP_TOKEN, without 'Bearer ' inside the variable. The Cursor JSON references ${env:BR_MCP_TOKEN}; Codex and Claude Code also read BR_MCP_TOKEN from the environment. Make sure your client inherits that environment, then confirm the server URL is https://brain-researcher.com/mcp and the Accept header is 'application/json, text/event-stream'. Restart the client, then ask it to call server_info (expect ok=true) and system_self_test (expect overall=pass). If it still can't see tools, have it inspect the exposed MCP tool names before trying anything else.
How do I verify my MCP connection is working?
Ask your agent to run the two health-check tools: server_info should return ok=true, and system_self_test should return overall=pass. The /mcp/setup page has ready-made smoke prompts for Codex and Claude Code. Always run these before claiming the server is connected or before requesting a workflow recipe.
Where do my run results go?
Use MCP to inspect persisted runs and their artifacts: run_list finds run IDs, run_get returns the status and step records, run_logs reads log payloads, and artifact_list enumerates the produced files. Keep the run ID with the result so its evidence and manifest stay traceable.
Does get_execution_recipe actually run the analysis?
No. get_execution_recipe returns a RECIPE - the exact command, required inputs, expected artifacts, and any blockers - not a completed analysis. This is BR's 'execution boundary': preparing a plan is not proof it ran. Execution only counts as successful when the expected artifacts, logs, and run manifest are actually produced and inspected. The recipe is what you (or your agent) then run locally, in a container, on Neurodesk, or on a cluster.
Can I run BR's MCP server locally or air-gapped?
Yes. Instead of the hosted cloud URL, run the server locally with: npx -y @brain-researcher/mcp-server start. For local-only data processing or air-gapped use, set ALLOW_NETWORK=false and ALLOWED_ROOTS=/data so the server only touches the paths you allow. This is the 'Local (Advanced)' tab on /mcp/setup.
How does BR handle my data?
With the hosted cloud MCP, requests go to brain-researcher.com over an authenticated connection tied to your personal token. If you need data to stay on your own machine, use the local MCP server (npx -y @brain-researcher/mcp-server start) with ALLOW_NETWORK=false and ALLOWED_ROOTS scoped to the directories you permit, so the server never reaches the network and only reads the roots you list. Tokens are personal, one active per user, and can be rotated or revoked at any time on /mcp/setup.
What does 'claim record' or 'grounding' mean in BR?
BR's core philosophy: a generated output is NOT a scientific claim by default. A claim record turns researcher judgment into executable commitments - the allowed alternatives, validation rules, provenance, and the boundaries beyond which a result shouldn't be read. 'Grounding' means tying a statement to real BR evidence (a dataset, workflow, KG node, or run) rather than asserting it. Every result stays linked to the evidence behind it and to its stated limits.
How do I cite a BR result or claim?
Cite the result through its provenance, not as a bare statement. Each run and analysis keeps its evidence trail (dataset + version, workflow, parameters, run manifest, and KG evidence), so reference the specific run/analysis in Vault and the datasets/workflows it grounds in. Because outputs aren't claims by default, report the verified claim together with its boundaries and validation rules rather than the raw output alone.