Trouble interacting with the ConPort

[bkutasi]

I have some trouble interacting ConPort with or without the RooFlow instructions using Gemini 2.5 pro preview 05-06. I did the setup with the exported system prompt, then run the conport specific installer.
The problem is that the memory is not retrieved or only sometimes(when manually asked) it retrieves specifics but mostly the ones I'm asking for.

"Initialize according to custom instructions" often leads to the model trying to get the info from the server with get_product_context but outright fails because it uses the wrong arguments. Then the failure concludes the subtask:

[CONPORT_INACTIVE] The attempt to retrieve the product context from ConPort failed with a validation error. The server-side tool get_product_context seems to expect arguments under a specific key (raw_args_from_fastmcp) which was not provided in the initial call structure.

As per the task instructions, if any get_* call fails unexpectedly, ConPort memory will not be used for this session.

ConPort memory will not be used for this session. Status: [CONPORT_INACTIVE].

I tried removing the custom instructions and just using the Flow-Architect mode since it already includes the MCP details, but when told Initialize ConPort Memory Strategy it starts to go through all the possible uses of conport, but obviously it's empty I just wanted to initialize it. The projectBrief.md is in my root.
Am i missing something? Maybe its a flow or model problem?

[GreatScottyMac]

The mentioning of raw_args_from_fastmcp means that there are either old rooflow instructions or old python cache present in the system. Reinstall rooflow (the conport custom instructions were updated recently) and then delete the context-portal/ MCP directory. Then run a pyclean type of command or some other method to clear the pycache from wherever necessary. If you're unsure or not familiar with clearing python cache, get instructions from an LLM, preferably one that's familiar with your system. Then reinstall comport mcp and you should be good to go. Please let me know if this works!

…

On Fri, May 16, 2025 at 6:28 AM bkutasi ***@***.***> wrote: *bkutasi* created an issue (GreatScottyMac/RooFlow#64) <#64> I have some trouble interacting ConPort with or without the RooFlow instructions using Gemini 2.5 pro preview 05-06. I did the setup with the exported system prompt, then run the conport specific installer. The problem is that the memory is not retrieved or only sometimes(when manually asked) it retrieves specifics but mostly the ones I'm asking for. "Initialize according to custom instructions" often leads to the model trying to get the info from the server with get_product_context but outright fails because it uses the wrong arguments. Then the failure concludes the subtask: [CONPORT_INACTIVE] The attempt to retrieve the product context from ConPort failed with a validation error. The server-side tool get_product_context seems to expect arguments under a specific key (raw_args_from_fastmcp) which was not provided in the initial call structure. As per the task instructions, if any get_* call fails unexpectedly, ConPort memory will not be used for this session. ConPort memory will not be used for this session. Status: [CONPORT_INACTIVE]. I tried removing the custom instructions and just using the Flow-Architect mode since it already includes the MCP details, but when told Initialize ConPort Memory Strategy it starts to go through all the possible uses of conport, but obviously it's empty I just wanted to initialize it. The projectBrief.md is in my root. Am i missing something? Maybe its a flow or model problem? — Reply to this email directly, view it on GitHub <#64>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BNSRHLJ4QW6BFQSRQHLNE7L26W4WTAVCNFSM6AAAAAB5IMXJQGVHI2DSMVQWIX3LMV43ASLTON2WKOZTGA3DQNRRGYYDAMI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[edxeth]

@GreatScottyMac flow-orchestrator unlike the old Boomerang has no memory bank knowledge. I'm using conport MCP and the only way to use it is if I select other Flow roles manually which defeats the purpose of Boomerang tasks, making flow-orchestrator useless unless I'm getting it wrong. Back when we used the custom Boomerang role and file based memory management the role was clever enough to init and update memory bank, but now no more.
I've also noticed that the memory bank does not update automatically when a task is completed unless specifically mentioned. I think that happens because I'm not using flow-orchestrator since it does not work as intended.

The steps above do not work with flow-orchestrator at all.

[GreatScottyMac]

I didn't attach the comport instructions to orchestrator because I wasn't sure how it would handle that. If you want to give it a try and let me know, that would be great. Just paste the contents of roo_code_conport_strategy <https://github.com/GreatScottyMac/RooFlow/blob/main/config/roo_code_conport_strategy> to the end of your system-prompt-flow-orchestrator <https://github.com/GreatScottyMac/RooFlow/blob/main/config/.roo/system-prompt-flow-orchestrator> .

…

On Sat, May 17, 2025 at 7:26 AM edx.eth ***@***.***> wrote: *edxeth* left a comment (GreatScottyMac/RooFlow#64) <#64 (comment)> @GreatScottyMac <https://github.com/GreatScottyMac> flow-orchestrator unlike the old Boomerang has no memory bank knowledge. I'm using conport MCP and the only way to use it is if I select other Flow roles manually which defeats the purpose of Boomerang tasks, making flow-orchestrator meaningless unless I'm getting it wrong. Back when we used the custom Boomerang role and file based memory management the role was clever enough to init and update memory bank, but now no more. — Reply to this email directly, view it on GitHub <#64 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BNSRHLIQFZFNF3WW6XSJEKD264MHJAVCNFSM6AAAAAB5IMXJQGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDQOBYGMYTGOBWGY> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[wokkieman]

roodmode also needs read and MCP access. But once that is in place, I didn't get orchestrator to actually read conport.

edit: the tools available for orchestrator also need to be updated

[GreatScottyMac]

The ConPort custom instructions aren't included in orchestrator mode but you can manually paste them to the end of its system prompt.

…

On Mon, May 19, 2025, 6:15 PM RJ ***@***.***> wrote: *wokkieman* left a comment (GreatScottyMac/RooFlow#64) <#64 (comment)> roodmode also needs read and MCP access. But once that is in place, I didn't get orchestrator to actually read conport. — Reply to this email directly, view it on GitHub <#64 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BNSRHLKWXASDXPBA4FJ6INT27JJZJAVCNFSM6AAAAAB5IMXJQGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDQOJSGQYDSNZYG4> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[wokkieman]

The ConPort custom instructions aren't included in orchestrator mode but
you can manually paste them to the end of its system prompt.
…

yes, I pasted them in as you said, but it's not enough. Roomodes need to be updated to give read and mcp group access. Once done, the next change would be the "Tools Definition Section" in orchestrator system prompt. Otherwise Roo gives this answer
I'm trying to access the ConPort data but I'm having trouble finding the right tool to use. From the tools available in flow-orchestrator mode (ask_followup_question, attempt_completion, switch_mode, new_task), which one should I use to interact with ConPort?

The architect symtem prompt has things like this
mcp_access: description: | Access to connected MCP servers providing additional tools and resources. Each server offers different capabilities to enhance task accomplishment. tools: - name: "use_mcp_tool" description: "Execute a specific tool provided by a connected MCP server." - name: "access_mcp_resource" description: "Access data or resources provided by a connected MCP server via URI."

` - name: use_mcp_tool
description: |
Executes a specific tool provided by a connected MCP (Multi-Capability Provider) server.
MCP servers offer additional capabilities and tools with defined input schemas.
Use this to leverage specialized functionalities offered by external servers (e.g., weather forecasts, database queries, external APIs).
parameters:
- name: server_name
required: true
description: The unique name identifying the connected MCP server that provides the desired tool.
- name: tool_name
required: true
description: The name of the specific tool to execute on the designated MCP server.
- name: arguments
required: true
description: |
A JSON object containing the input parameters for the tool.
This object MUST strictly adhere to the input schema defined by the specific tool being called on the MCP server.
Ensure all required parameters are included and data types match the schema.
usage_format: |
<use_mcp_tool>
<server_name>[MCP server name here]</server_name>
<tool_name>[Tool name on that server]</tool_name>

{
"param1": "value1",
"param2": 123,
... # Ensure this JSON matches the tool's schema
}

</use_mcp_tool>
example:
- description: Request a 5-day weather forecast for San Francisco from the 'weather-server' MCP
usage: |
<use_mcp_tool>
<server_name>weather-server</server_name>
<tool_name>get_forecast</tool_name>

{
"city": "San Francisco",
"days": 5
}

</use_mcp_tool>
- description: Request user details from the 'auth-server' MCP using a user ID
usage: |
<use_mcp_tool>
<server_name>auth-server</server_name>
<tool_name>get_user_details</tool_name>

{
"user_id": "usr_1a2b3c"
}

</use_mcp_tool> # Added another example for variety

• name: access_mcp_resource
  description: |
  Accesses or retrieves data from a specific resource provided by a connected MCP (Multi-Capability Provider) server.
  Resources represent data sources that can be used as context, such as files, API responses, database tables, or system information, identified by a unique URI.
  Use this to fetch context or data from external systems managed by MCP servers.
  parameters:
  • name: server_name
    required: true
    description: The unique name identifying the connected MCP server that provides the desired resource.
  • name: uri
    required: true
    description: |
    The Uniform Resource Identifier (URI) that uniquely identifies the specific resource to be accessed on the designated MCP server.
    The format of the URI depends on the specific MCP server and the resource type it provides.
    usage_format: |
    <access_mcp_resource>
    <server_name>[MCP server name here]</server_name>
    [Unique resource URI here]
    </access_mcp_resource>
    example:
  • description: Access the current weather conditions for San Francisco from the 'weather-server' MCP
    usage: |
    <access_mcp_resource>
    <server_name>weather-server</server_name>
    weather://san-francisco/current
    </access_mcp_resource>
  • description: Access the latest system log file from the 'monitoring-server' MCP
    usage: |
    <access_mcp_resource>
    <server_name>monitoring-server</server_name>
    logs://system/latest
    </access_mcp_resource> # Added another example for variety
  • description: Access a specific database record from the 'database-server' MCP
    usage: |
    <access_mcp_resource>
    <server_name>database-server</server_name>
    db://users/id/12345
    </access_mcp_resource> # Added another example for variety`

[GreatScottyMac]

I gave orchestrator the same tools that it has in its default Roo Code system prompt. How does the default orchestrator handle this type of thing? You could try building a fully capable system prompt for it by combining rooflow_core_prompt.yaml and roo_code_conport_strategy?

If you want to try that, let me know how it works.

[GreatScottyMac]

Flow-orchestrator mode now has the tools and instructions needed to interact with either roo-code-memory-bank or context-portal.

[edxeth]

Flow-orchestrator mode now has the tools and instructions needed to interact with either roo-code-memory-bank or context-portal.

Good job, thanks for the amazing update!