OppenSteward-MCP
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| OPPEN_HOST | No | Host interface for HTTP mode. Default is 127.0.0.1. | 127.0.0.1 |
| OPPEN_PORT | No | Local port for HTTP mode. Default is 8766. | 8766 |
| OPPEN_STATE_DIR | No | Directory for runtime state (login token, OAuth DB, logs). Default is .runtime. | .runtime |
| OPPEN_TRANSPORT | No | Transport mode. Use 'stdio' for Secure MCP Tunnel or 'http' for domain forwarding with OAuth. Defaults to 'stdio' for new installations. | stdio |
| OPPEN_TUNNEL_ID | No | Tunnel ID from OpenAI Platform, required when using Secure MCP Tunnel. | |
| OPPEN_PUBLIC_URL | No | Public HTTPS URL used in HTTP + OAuth mode (e.g., https://projects.example.com). Required when using HTTP transport. | |
| OPPEN_SCAN_ROOTS | No | JSON array of directories to scan for projects, e.g., ["~/Projects", "~/Research"]. Defaults to the current user's home directory. | |
| OPPEN_SKILL_ROOT | No | Directory containing the skill folders 'oppen-project-steward' and 'stepwise-r-project'. If not set, common locations like ~/.codex/skills and ~/.agents/skills are tried. | |
| OPPEN_SCAN_SECONDS | No | Max seconds per scan batch. Default is 90. | 90 |
| OPPEN_EXCLUDE_ROOTS | No | JSON array of directories to exclude from scanning, e.g., ["~/Projects/private"]. | |
| OPPEN_MAX_SCAN_DIRS | No | Max directories to scan per batch. Default is 500000. | 500000 |
| OPPEN_SCAN_INTERVAL | No | Seconds between scan cycles; minimum 10. Default is 300. | 300 |
| OPPEN_TUNNEL_CLIENT | No | Full path to the tunnel-client executable if it is not on your PATH. | |
| OPPEN_TUNNEL_PROFILE | No | Profile name used for the tunnel-client. Default is 'oppen-steward'. | oppen-steward |
| CONTROL_PLANE_API_KEY | No | Runtime key from OpenAI Platform for the tunnel, required when using Secure MCP Tunnel. | |
| OPPEN_DISCUSSION_MODE | No | Discussion feature mode: 'off' (default), 'read' to allow reading discussion files, or 'write' to allow reading and creating/editing discussions. | off |
| OPPEN_EXTRA_REDIRECT_URIS | No | JSON array of extra allowed OAuth redirect URIs. Default is []. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| list_projectsA | List explicitly registered projects, stable IDs, roots and skill versions. No disk scanning. |
| refresh_projectsA | Reload the project list and recheck its exact roots. Never scan or register other projects. |
| project_overviewC | Read the registry, locate indices and report this connection's Discussion permissions. discussion_mode describes local configuration; discussion_access describes authorization. mcp_tools is the live server list, which may differ from an older client's cached tool list. |
| list_filesA | Browse registry and indexed Memory/Attention paths. Use list_discussions for Discussion. |
| read_fileA | Read an allowlisted governance document in chunks. Offsets/lengths are BYTES. Follow next_offset to EOF; compare size/modified_ns across chunks and restart if changed. Maximum chunk 262144 bytes; base64 preserves exact bytes of allowed governance text. Data, source, results, Audit, README and other canonical bodies are never available. Discussion uses read_discussion and its separate permission; this tool cannot read Discussion. |
| searchA | Search only allowlisted governance documents across projects; returns citation IDs for fetch. Content search covers UTF-8 files up to 256 KiB. Narrow project_id/glob if truncated. |
| fetchB | Fetch a search result by ID. Follow metadata.next_offset to read a long document completely. |
| get_skill_guideB | Read the locally installed SKILL.md for either supported management skill. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 8 tools
The three listing/introspection tools (list_files, list_projects, project_overview) overlap substantially—all touch the project registry and indexed paths—and fetch vs read_file both retrieve document content by different keying schemes. Detailed descriptions and cross-references help, but an agent could still misselect among the registry-related tools.
Names are consistently snake_case with a mostly verb_noun pattern (list_files, list_projects, refresh_projects, get_skill_guide, read_file). Minor deviations: 'search' and 'fetch' are bare verbs, and 'project_overview' is noun-style, but the set remains predictable.
Eight tools is well-scoped for a read-only governance/registry browsing server. Each tool maps to a distinct capability (enumerate projects, browse paths, search, retrieve, read chunks, load skill docs) without obvious padding.
The surface is read-only and coherent, but the descriptions repeatedly reference list_discussions and read_discussion—tools that are not included in this set—creating dead-end pointers and a notable gap in the Discussion workflow it claims to support. Core search/fetch/read lifecycle is otherwise covered.