Qodo MCP server: Context Engine, formerly Aware
Why searches for Qodo Aware lead to a dead host, what the two agents behind the connector actually do, and which facts Qodo does not publish.
Last verified 25 August 2026 · from docs.qodo.ai, the npm registry, DNS and the GitHub API
This page is one of 90 in a directory of monitoring and developer tools. Each page checks what an AI coding agent can get out of the tool through its MCP server, the connector that lets the agent query the tool directly.
Summary
The Qodo MCP server has a new name: Qodo Aware is now Context Engine MCP. Qodo's own documentation states the rename, and the old host open-aware.qodo.ai no longer resolves at all: DNS returns NXDOMAIN.
The predecessor is still on GitHub. qodo-ai/open-aware has 503 stars and was last pushed on 29 October 2025, so search results still point at a project that has moved.
The connector is a deep research agent over your indexed repositories, not a set of API wrappers. Qodo describes indexing across as many as thousands of repositories, then agentic reasoning over what it retrieves.
No tool names or counts are published, there is no MCP registry entry, and no installable package exists, so this page claims no inventory.
What is the Qodo MCP server?
Most connectors in this directory wrap an API: a tool per endpoint, with typed parameters. Qodo's does something different. It indexes your repositories and answers questions about them.
Two agents sit behind it. Qodo names an Ask Agent for quick understanding and a Deep Research Agent for planning refactors, architectural changes and debugging across systems.
The documented pipeline is four steps: index the repositories, retrieve relevant functions, docs, commit history and patterns, reason over them with specialized agents, then generate an answer grounded in that.
Qodo contrasts this with plain retrieval. It says traditional systems retrieve a few snippets and reason only over those. If the right context is not pulled at the start, the answer misses.
Three names, one of which still works:
| Name | What it is | State |
|---|---|---|
| Context Engine MCP | The current product, documented under developer tools | Live in Qodo's 2.x documentation |
| Qodo Aware | The former name of the same thing | open-aware.qodo.ai returns NXDOMAIN |
qodo-ai/open-aware | The predecessor repository, still public | 503 stars, last pushed 29 October 2025 |
@qodo/command on npm is a different thing again: a client, and its latest version 0.36.0 was published on 11 January 2026 carrying a deprecation notice.
It sits squarely in code, tests and review, and reaches nothing operational.
| Part of the work | What Qodo has here | Can an agent reach it |
|---|---|---|
| Code, tests & review | Cross-repository code understanding, dependencies and impact | Yes. Deep. Qodo's own examples are where a function is used across repositories, and what breaks if an API response changes |
| CI & release | Commit history, as indexed context | Partial. Indirect. History informs answers, and nothing models a build, a pipeline or a deploy |
| Production observability | None | No metrics, logs or traces |
| Agent observability & evals | None | No traces or evaluations of model behavior |
| AI cost management | None | No spend data of any kind |
In practice, Qodo answers questions about the shape of a codebase that no per-endpoint connector can. Examples: what calls this, what breaks if it changes, how a feature works end to end. It answers nothing about what the code did in production.
What can an AI agent do with Qodo?
No tool names or counts are published. Qodo documents Context Engine MCP as a deep research agent for large codebases, connecting to Cursor, Windsurf, Claude Desktop and GitHub Copilot, and stops short of listing what it exposes. What it does describe is the shape of the work: cross-repository code understanding and impact analysis. Get the tool list from Qodo directly.
Qodo MCP server limits: where answers come back incomplete
Qodo's first limit is a search result. The old name still ranks, and its host no longer resolves.
The old name still ranks, and its host is gone.
Qodo Aware was renamed Context Engine MCP. The host open-aware.qodo.ai now returns NXDOMAIN, so a link from an article or a bookmark fails with a DNS error rather than a redirect.
qodo-ai/open-aware remains public with 503 stars, last pushed 29 October 2025, which keeps the old name visible in search.
Guard: Search for Context Engine MCP. Treat anything naming Aware as predating the rename.
There is no published tool list, and no way to capture one.
Qodo publishes no tool names or counts, there is no registry entry, and no package exists to run locally. The connector is a hosted service reached from a client.
Every other measurement on this site starts from a captured tools/list. Here there is nothing to capture without an account.
Guard: Measure the connection yourself once configured, rather than planning a context budget from published figures that do not exist.
The agent says: “Nothing calls that function.” Nothing in the index does.
Answers come from an index of the connected repositories. A function added or renamed since the last index will not appear.
Guard: Have the agent state that the answer came from an index, as the disclosure block above requires.
The agent says: “That service does not exist.” Its repository may not be connected.
Qodo states it analyzes only repositories that have been explicitly connected, and can reason only about those.
Guard: Check the connected repository list before treating an absence as a finding.
How to configure Qodo MCP for agents
- 1
Connect the repositories deliberately
Qodo states it analyzes only repositories that have been explicitly connected, and can reason about nothing else.
- 2
Choose the deployment
Cloud, hybrid, or an MCP server for air-gapped and self-managed environments.
- 3
Pick the agent for the question
Ask Agent for quick understanding, Deep Research Agent for refactors, architectural changes and cross-system debugging.
- 4
Scope answers to repositories where it matters
Qodo documents scoping as a supported control on a query.
- 5
Expect no tool inventory
Nothing is published, so budget context by measuring the connection rather than reading a table.
Paste this into your agent’s instructions
An answer here is a research result over an index, not a record read from an API. How current the index is decides whether the answer is right, and nothing in the prose reveals that.
When you answer using Qodo Context Engine MCP, state: - Which repositories the answer is scoped to. It can reason only about repositories that are explicitly connected. - Whether you used the quick Ask path or the Deep Research path. - That the answer comes from an index of the codebase, so a very recent change may not be reflected in it. Never state that data does not exist. State the repositories you searched and what came back. Write your answer in ASD-STE100 Simplified Technical English. Use short sentences with one idea in each.
Do you need the MCP server at all?
Only for cross-repository questions. GitHub and GitLab both publish servers that read code directly, which is cheaper when you already know which file to open.
Ripgrep and a language server answer "where is this used" faster than any indexed service, provided the code is on the machine.
For review rather than research, CodeRabbit, SonarQube and Semgrep analyze a diff instead of a codebase.
Qodo's own IDE plugin and agent skills cover the same engine inside an editor, which is a different surface from the connector.
What MCP adds is cross-repository reasoning that a per-file tool cannot do: what breaks if this API changes, across services nobody has open.
Use it for questions that span repositories. Use a code-reading server for questions that span one file.
Qodo MCP server setup
Qodo's connector is official and was renamed. It is now Context Engine MCP, described as a deep research agent for large codebases exposed to AI tools over the protocol.
One thing shapes this whole section: Qodo publishes no client configuration. The documentation names deployment models and supported clients, and stops short of a config block or a tool list.
So the honest route is Qodo's own onboarding, and the checks below are what you can confirm from outside it.
Choose the deployment model first
Qodo documents three: a hosted service, a hybrid deployment, and a self-managed server for environments including air-gapped ones. The model decides where the engine indexes your code.
No credentials appear here because no configuration is published to put them in. Provisioning runs through Qodo rather than through a file you edit.
cloud · hybrid · self-managed (air-gapped supported)
Confirm your client is supported
The documentation names Cursor, Windsurf, Claude Desktop and GitHub Copilot. If yours is not among them, that is worth resolving with Qodo before any rollout.
This is a compatibility check and involves no credentials.
Cursor · Windsurf · Claude Desktop · GitHub Copilot
Every other client
Each block below is the configuration for one client, with the file path and the key that client expects.
What could not be verified from outside
There is no public package and no registry entry. @qodo/context-engine-mcp and qodo-aware both return 404 from npm, and a registry search for qodo on 25 August 2026 returned zero results.
The predecessor host open-aware.qodo.ai returns NXDOMAIN, which is consistent with the rename rather than with a withdrawal. These checks need no credentials, and no auth is involved.
npm view @qodo/context-engine-mcp # 404 host open-aware.qodo.ai # NXDOMAIN
The predecessor repository is still there
qodo-ai/open-aware carries 503 stars and was last pushed on 29 October 2025. It is the earlier form of the same work and is not the current product.
Reading its metadata needs no credentials, and it is the clearest evidence that the rename is real rather than a rebrand of nothing.
qodo-ai/open-aware 503 stars pushed 29 Oct 2025
Read-only access, permissions and security
It indexes your source, which is the point and the exposure
Qodo states each organization's index is isolated and never mixed with other tenants.
Qodo states it does not train on customer code
Its documentation says the product does not use your code for training any foundation models.
Retrieval is documented as temporary
Qodo states relevant snippets are retrieved for context at query time and are not stored beyond session needs.
Encryption is stated as TLS in transit and AES-256 at rest
No third-party sharing of code or metadata is stated.
Air-gapped deployment is offered
Qodo names self-managed and air-gapped environments as a supported deployment for the MCP server, which is the option where none of the above has to be taken on trust.
Access is scoped by what you connect
No repository is reachable without configuration, and role-based access is documented as an enterprise control.
Troubleshooting
- `open-aware.qodo.ai` does not resolve
- Correct, and it is not a network problem. DNS returns NXDOMAIN for that host. The product is now Context Engine MCP, documented under developer tools.
- You cannot find it in the MCP registry
- It is not there. A registry search for qodo returned zero results on 25 August 2026, and there is no installable package for it.
- `@qodo/command` warns that it is deprecated
- That package is a client, not this server. Version 0.36.0, published 11 January 2026, carries the notice that the package is no longer supported.
- Answers miss a change you just merged
- The connector answers from an index. A change lands in an answer once indexing has caught up with it, which is why the disclosure block above makes the agent say so.
- It cannot see a repository
- Only explicitly connected repositories are analyzed. Qodo states no code is accessed without configuration.
Qodo MCP server: Reference
| Item | Value |
|---|---|
| Current name | Qodo Context Engine MCP |
| Former name | Qodo Aware |
| Old host | open-aware.qodo.ai, NXDOMAIN on 25 August 2026 |
| Predecessor repository | qodo-ai/open-aware · 503 stars · pushed 29 October 2025 |
| MCP registry entry | None |
| Installable package | None |
| Tools | Not published |
| Agents | Ask Agent and Deep Research Agent |
| Deployment | Cloud, hybrid, or self-managed for air-gapped environments |
| Clients named | Cursor, Windsurf, Claude Desktop and GitHub Copilot |
| Training on customer code | Qodo states none |
| Inventory source | docs.qodo.ai Context Engine MCP overview, read 25 August 2026 |
What the vendor documents
The predecessor repository carries 503 stars, and public discussion of the renamed connector is thin. The useful record is what the rename left behind. The counts below are that record.
| What was checked | What it shows |
|---|---|
| Predecessor repository stars | 503 |
| Its last push | 29 October 2025 |
| Old host DNS | NXDOMAIN |
| MCP registry entries | 0 |
| Installable packages | 0 |
| Published tool count | None |
| Agents documented | 2 |
“Qodo Context Engine MCP, formerly Qodo Aware, provides multi-repo intelligence via the Model Context Protocol.”
“Qodo Context Engine MCP only analyzes repositories that have been explicitly connected. No code is accessed without configuration.”
“Traditional RAG systems retrieve a few snippets and reason only over those. If the right context isn't pulled at the start, the answer can miss the mark.”
The MCP registry was searched for qodo on 25 August 2026 and returned zero results. @qodo/context-engine-mcp and qodo-aware both return 404 from the npm registry. open-aware.qodo.ai returns NXDOMAIN. Repository facts come from the GitHub API on the same date.
Should you connect an agent to Qodo?
Yes. Best for cross-repository impact analysis, scoped to the repositories you connected.
- Strongest fit
- Teams using Qodo's codebase index to ask cross-repository impact questions, such as what breaks if this contract changes.
- Main advantage
- It reasons over an index of the whole codebase, which is a different capability from reading files on demand.
- Main weakness
- Nothing about the tool list is published, so it cannot be sized or budgeted before it is connected.
- Operational risk
- Answers come from an index, so a recent change can be missing without anything in the answer saying so.
From Oodle
One platform for agent traces and infrastructure
Agent traces usually sit in a different product from the rest of your telemetry, so when a slow database makes an agent slow the symptom is in one tool and the cause is in another. Oodle keeps both in one query surface, from $10 per million spans.
See agent observabilityFrequently asked questions
What happened to Qodo Aware?
It was renamed Context Engine MCP. The old host open-aware.qodo.ai no longer resolves, returning NXDOMAIN, though qodo-ai/open-aware is still on GitHub with 503 stars.
How many tools does it expose?
Qodo does not publish tool names or counts, there is no registry entry, and no installable package exists, so this page claims no inventory.
What is the difference between the two agents?
Qodo documents an Ask Agent for quick understanding and a Deep Research Agent for planning refactors, architectural changes and debugging across systems.
Does it train on my code?
Qodo states it does not use customer code for training any foundation models, that retrieval is temporary, and that each organization's index is isolated.
Can it run somewhere isolated?
Yes. Qodo names cloud, hybrid and self-managed deployments, with the MCP server route offered specifically for air-gapped or self-managed environments.
Is `@qodo/command` the same thing?
No. That is a client package, and its latest version 0.36.0, published 11 January 2026, carries a deprecation notice saying it is no longer supported.
