Install and Use GBrain MCP in Zoo Code

1. Installation

Open Zoo Code → click ⚙️ → choose Edit Global MCP (for all projects) or Edit Project MCP (saved in .roo/mcp.json).

Paste the following config into the file:

{
  "mcpServers": {
    "gbrain": {
      "type": "streamable-http",
      "url": "http://gbrain:7333/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN"
      }
    }
  }
}

Note: If the server uses SSE (legacy), change "type" to "sse". Try streamable-http first.


2. Connection check

2a. In the Zoo Code UI

  • Click ⚙️ → view the list of MCP servers
  • Green dot = connection OK
  • Red dot = error → hover to see the message
curl -X POST http://gbrain:7333/mcp \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'
  • Returns JSON with the tool list → server OK
  • Error 401 → wrong token
  • Connection refused → server not running or wrong host/port

2c. View detailed logs

Use View → Output → select “Zoo Code - MCP” to see full connection logs.


3. Optimize — reduce from 85 to under 60 tools

Zoo Code warns when there are too many tools (>60) because that can confuse the model.

Keep about 38 tools

Group Tools
Search / query search, query, think
Page get_page, put_page, list_pages, delete_page
Facts / memory recall, extract_facts, find_contradictions
Code intelligence code_def, code_refs, code_callers, code_flow
Health / diagnostics get_health, run_doctor, get_stats, advisor
Graph get_links, get_backlinks, traverse_graph, add_link, remove_link
Skills list_skills, get_skill
Jobs (basic) submit_job, get_job, get_job_progress, list_jobs
Identity whoami, get_brain_identity, get_status_snapshot
Tags / data add_tag, remove_tag, get_tags, get_chunks
Timeline add_timeline_entry, get_timeline
Experts find_experts, get_recent_salience

Disable the less important tools

Add this to the config:

{
  "mcpServers": {
    "gbrain": {
      "type": "streamable-http",
      "url": "http://gbrain:7333/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN"
      },
      "disabledTools": [
        "schema_lint", "schema_graph", "schema_explain_type",
        "schema_review_orphans", "schema_apply_mutations",
        "reload_schema_pack", "run_onboard", "get_active_schema_pack",
        "list_schema_packs", "schema_stats",
        "takes_list", "takes_search", "takes_scorecard",
        "takes_calibration", "get_calibration_profile",
        "cancel_job", "retry_job", "pause_job", "resume_job",
        "replay_job", "send_job_message", "submit_agent",
        "sources_add", "sources_list", "sources_remove", "sources_status",
        "get_versions", "revert_version", "restore_page",
        "log_ingest", "get_ingest_log",
        "find_trajectory", "find_anomalies", "find_orphans",
        "volunteer_context", "search_by_image",
        "list_link_sources", "run_skillopt", "list_brain_skillpack",
        "code_blast", "code_callees",
        "resolve_slugs", "put_raw_data", "get_raw_data",
        "run_doctor"
      ]
    }
  }
}

4. Using it in Zoo Code

Zoo Code automatically detects the best tool when you chat — you do not need to invoke tools manually.

"Find information about X in gbrain"
"Ask gbrain about Y"
"Search gbrain for Z"

Read / write content

"Get page [name] from gbrain"
"Update page [name] with the following content..."
"List pages in gbrain"

Code analysis

"Who calls function X in gbrain?"
"Where is function Y defined?"
"Show the call flow for Z"

System checks

"Is gbrain healthy?"
"Show gbrain stats"
"Run doctor check on gbrain"

Better prompts when Zoo Code does not choose a tool automatically

"Use gbrain to find..."
"Call the get_page tool in gbrain with slug = ..."

Quick reference

Purpose Tool
Smart query / search search, query, think
Read/write knowledge get_page, put_page, list_pages
Relationship graph get_backlinks, traverse_graph
Code analysis code_def, code_refs, code_flow
System operation get_health, get_stats, advisor
Background tasks submit_job, get_job_progress
Memory / facts recall, extract_facts